control solutions for multiphase flow undertittel på ... pipeline-riser system. the new model, and...
TRANSCRIPT
ISBN 978-82-471-xxxx-x (printed version)ISBN 978-82-471-xxxx-x (electronic version)
ISSN 1503-8181
Doctoral theses at NTNU, 2010:XX
Fornavn Etternavn
Doctoral theses at NTNU, 2010:23
Fornavn Etternavn
NTNUNorwegian University of Science and Technology
Thesis for the degree of philosophiae doctorFaculty of Engineering Science and Technology
Department of Marine Technology
Tittel på avhandlingen
Undertittel på avhandlingen
Doctoral theses at NTNU, 2013:271
Esmaeil Jahanshahi
Control Solutions for Multiphase Flow
Linear and nonlinear approaches to anti-slug control
ISBN 978-82-471-4670-5 (printed version)ISBN 978-82-471-4671-2 (electronic version)
ISSN 1503-8181
Doctoral theses at N
TNU
, 2013:271Esm
aeil Jahanshahi
NTN
UN
orw
egia
n U
nive
rsity
of S
cien
ce a
nd T
echn
olog
yTh
esis
for
the
degr
ee o
f Phi
loso
phia
e D
octo
rFa
cult
y of
Nat
ural
Sci
ence
s an
d Te
chno
logy
Dep
artm
ent o
f Che
mic
al E
ngin
eeri
ng
Thesis for the degree of philosophiae doctor
Trondheim, xxxx 2010
Norwegian University of Science and TechnologyFaculty of Engineering Science and TechnologyDepartment of Marine Technology
Fornavn Etternavn
Tittel på avhandlingenUndertittel på avhandlingen
Department of Marine Technology
Esmaeil Jahanshahi
Control Solutions for Multiphase FlowLinear and nonlinear approaches to anti-slug control
Thesis for the degree of Philosophiae Doctor
Trondheim, October 2013
Norwegian University of Science and TechnologyFaculty of Natural Sciences and TechnologyDepartment of Chemical Engineering
NTNUNorwegian University of Science and Technology
©
ISSN 1503-8181
IMT Report 2010-xx
Doctoral Theses at NTNU, 2010:xx
Printed by Skipnes Kommunikasjon as
Thesis for the degree of philosophiae doctor
Faculty of Engineering Science and TechnologyDepartment of Marine Technology
Fornavn Etternavn
ISBN 82-471-xxxx-x (printed ver.)ISBN 82-471-xxxx-x (electronic ver.)
NTNUNorwegian University of Science and Technology
Thesis for the degree of Philosophiae Doctor
Faculty of Natural Sciences and TechnologyDepartment of Chemical Engineering
© Esmaeil Jahanshahi
ISBN 978-82-471-4670-5 (printed version)ISBN 978-82-471-4671-2 (electronic version)ISSN 1503-8181
Doctoral theses at NTNU, 2013:271
Printed by Skipnes Kommunikasjon as
SUMMARY
Severe-slugging flow in offshore production flowlines and risers is undesirableand effective solutions are needed to prevent it. Automatic control usingthe top-side choke valve is a recommended anti-slug solution, but manyanti-slug control systems are not robust against plant changes and inflowdisturbances. The closed-loop system becomes unstable after some time,and the operators turn off the controller. In this thesis, the focus is onfinding robust anti-slug solutions using both linear and nonlinear controlapproaches. The study includes mathematical modeling, analysis, OLGAsimulations and experimental work.
First, a simplified dynamical four-state model was developed for a severe-slugging pipeline-riser system. The new model, and five other models ex-isting in the literature, were compared with results from the OLGA simu-lator. OLGA is a commercial package based on rigorous models and it iswidely used in the oil industry. The new four-state model is also verifiedexperimentally. Furthermore, the pipeline-riser model was extended to awell-pipeline-riser system by adding two new state variables.
Next, the simplified dynamical models were used for controllability anal-ysis of the system. From the controllability analysis, suitable controlledvariables and manipulated variables for stabilizing control were identified.A new mixed-sensitivity controllability analysis was introduced in which asingle γ-value quantifies the robust performance of the control structure. Inagreement with previous works, subsea pressure measurements were foundto be the best controlled variables for an anti-slug control. The top-sidevalve is usually used as the manipulated variable, and two alternative loca-tions also were considered. It was found that a subsea choke valve close tothe riser base has the same operability as the top-side choke valve, while awell-head valve is not suitable for anti-slug control.
Three linear control solutions were tested experimentally. First, H∞
control based on the four-state mechanistic model of the system was ap-plied. H∞ mixed-sensitivity design and H∞ loop-shaping design were con-
i
ii SUMMARY
sidered for this. Second, IMC design based on a identified model was chosen.The resulting IMC controller is a second-order controller that can be imple-mented as a PID controller with a low-pass filter (PID-F). Finally, PI-controlwas considered and PI tuning values were obtained from the proposed IMCcontroller. It was shown that the IMC (PID-F) controller has good per-formance and robustness, matching the model-based H∞ controller, and itdoes not need a mechanistic model and it is easier to tune.
Four nonlinear control solutions were tested; three of them are basedon the mechanistic model and the fourth one is based on identified mod-els. The first solution is state feedback with state variables estimated bynonlinear observers. Three types of observers were tested experimentallyand it was found that the nonlinear observers could be used only whenthe using the top-side pressure measurement. The second solution is anoutput-linearizing controller using two directly measured pressures. Thethird nonlinear controller is PI control with gain adaptation based on a sim-ple model of the static gain. The last nonlinear solution is a gain-schedulingof three IMC controllers which were designed based on identified models.The gain-scheduling IMC does not need a mechanistic model and showsbetter robustness compared to the other nonlinear control solutions.
ACKNOWLEDGEMENTS
This work could not have been completed without the assistance of severalindividuals. First and foremost is my supervisor, Prof. Sigurd Skogestad.Prof. Sigurd Skogestad has been a real source of motivation during thiswork. His doors were always open for suggestions and guidance and hispassion for learning has truly inspired me. He has definitely left a lastingimpression on me as a both a person and as a researcher. Next, I wouldreally like to thank my co-supervisor Prof. Ole Jørgen Nydal (Departmentof Energy and Process Engineering). He provided access to the multi-phaseflow laboratory and we always had fruitful discussion about slug flow.
Special thanks goes to Esten Ingar Grøtli (Department of EngineeringCybernetics) who introduced me to the world of Nonlinear Estimation andNonlinear Control, and was co-author for two of conference publications.I would also like to acknowledge with gratitude, the master students whocontributed to this project: Anette Helgesen, Knut Age Meland, Mats Lie-ungh, Henrik Hansen, Terese Syre, Mahnaz Esmaeilpour and Anne SofieNilsen.
I greatly appreciate help from the NTNU technical and laboratory staffwho helped to build the experimental rig for the anti-slug control, speciallyGeir Finnøy who implemented the electronics and the data acquisition sys-tem of the experimental rig.
I gratefully acknowledge the funding source that made this work possi-ble. This work has been financed by SIEMENS AS, Oil & Gas Solutionsthrough the project SIEMENS-NTNU Research Cooperation Oil and GasOffshore. I would like to thank the project manager from Siemens Dr.Anngjerd Pleym who helped in the timely completion of this work withassistance in the scheduling.
It was an absolute privilege to have the company of Process SystemsEngineering group members most of whom became my close friends andI will always cherish the tremendous warmth I received from them. They
iii
iv ACKNOWLEDGEMENTS
helped create an environment of sharing, as we learned together the lessonsof life as well as Process Control.
Finally, I would like to thank my parents, Rahmatollah and MahtalatJahanshahi who have always stood by me, even though living far away.
Esmaeil JahanshahiOctober 2013Trondheim, Norway
Contents
SUMMARY i
ACKNOWLEDGEMENTS iii
1 INTRODUCTION 1
1.1 Offshore Oil Production . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Petroleum reservoir . . . . . . . . . . . . . . . . . . . 2
1.1.2 Subsea oil wells . . . . . . . . . . . . . . . . . . . . . . 2
1.1.3 Subsea processing . . . . . . . . . . . . . . . . . . . . 3
1.1.4 Subsea pipelines and risers . . . . . . . . . . . . . . . 4
1.1.5 Topside separators . . . . . . . . . . . . . . . . . . . . 4
1.2 Flow Assurance Challenges . . . . . . . . . . . . . . . . . . . 6
1.2.1 Hydrates . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.2 Wax . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.3 Asphaltenes . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.4 Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.5 Corrosion . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.6 Emulsions . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.7 Slugging flow . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Riser Slugging . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.1 Mechanism of riser slugging . . . . . . . . . . . . . . . 13
1.3.2 Conventional anti-slug solutions . . . . . . . . . . . . . 14
1.3.3 Automatic control solution . . . . . . . . . . . . . . . 16
1.4 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Organization of Thesis . . . . . . . . . . . . . . . . . . . . . . 18
1.6 List of Contributions . . . . . . . . . . . . . . . . . . . . . . . 19
1.6.1 Conference publications . . . . . . . . . . . . . . . . . 19
1.6.2 Journal publications . . . . . . . . . . . . . . . . . . . 20
1.6.3 Patents . . . . . . . . . . . . . . . . . . . . . . . . . . 20
v
vi Contents
2 SIMPLIFIED DYNAMICAL MODELS 21
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 New simplified four-state model . . . . . . . . . . . . . . . . . 23
2.2.1 Mass balance equations for pipeline and riser . . . . . 23
2.2.2 Inflow conditions . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Outflow conditions . . . . . . . . . . . . . . . . . . . . 25
2.2.4 Pipeline model . . . . . . . . . . . . . . . . . . . . . . 25
2.2.5 Riser model . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.6 Gas flow model at riser base . . . . . . . . . . . . . . . 29
2.2.7 Liquid flow model at riser base . . . . . . . . . . . . . 29
2.2.8 Phase distribution model at outlet choke valve . . . . 29
2.3 Comparison of models . . . . . . . . . . . . . . . . . . . . . . 30
2.3.1 OLGA test case and reference model . . . . . . . . . . 30
2.3.2 Comparison of models with OLGA simulations . . . . 32
2.3.3 Comparison with experiments . . . . . . . . . . . . . . 35
2.4 Well-Pipeline-Riser System . . . . . . . . . . . . . . . . . . . 37
2.4.1 Simplified six-state model . . . . . . . . . . . . . . . . 37
2.4.2 Comparison to OLGA model . . . . . . . . . . . . . . 39
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3 CONTROLLED VARIABLE SELECTION 45
3.1 Controllability: Theoretical Background . . . . . . . . . . . . 46
3.1.1 Closed-loop transfer functions . . . . . . . . . . . . . . 46
3.1.2 Lower bound on S and T . . . . . . . . . . . . . . . . 47
3.1.3 Lower bound on KS . . . . . . . . . . . . . . . . . . . 48
3.1.4 Lower bounds on SG and SGd . . . . . . . . . . . . . 48
3.1.5 Lower bound on KSGd . . . . . . . . . . . . . . . . . 49
3.1.6 Pole vectors . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1.7 Mixed sensitivity controllability analysis . . . . . . . . 50
3.1.8 Low-frequency performance . . . . . . . . . . . . . . . 50
3.1.9 Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2 Well-Pipeline-Riser System . . . . . . . . . . . . . . . . . . . 51
3.2.1 Summary of simplified model . . . . . . . . . . . . . . 51
3.2.2 Bounds on minimum achievable peaks . . . . . . . . . 52
3.2.3 Control structure selection . . . . . . . . . . . . . . . . 54
3.3 Gas-Lifted Oil Well . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3.1 Summary of simplified model . . . . . . . . . . . . . . 58
3.3.2 Bounds on minimum achievable peaks . . . . . . . . . 61
3.3.3 Control structure selection . . . . . . . . . . . . . . . . 63
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Contents vii
4 MANIPULATED VARIABLE SELECTION 734.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734.2 Pipeline-Riser System . . . . . . . . . . . . . . . . . . . . . . 74
4.2.1 Experimental setup . . . . . . . . . . . . . . . . . . . . 744.2.2 OLGA model . . . . . . . . . . . . . . . . . . . . . . . 754.2.3 Simplified model . . . . . . . . . . . . . . . . . . . . . 76
4.3 Controllability Analysis . . . . . . . . . . . . . . . . . . . . . 794.4 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . 814.5 OLGA Simulation Results . . . . . . . . . . . . . . . . . . . . 824.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 864.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5 LINEAR CONTROL SOLUTIONS 895.1 Robust Control Based on Mechanistic Model . . . . . . . . . 89
5.1.1 H∞ mixed-sensitivity design . . . . . . . . . . . . . . 905.1.2 H∞ loop-shaping design . . . . . . . . . . . . . . . . . 91
5.2 IMC Based On Identified Model . . . . . . . . . . . . . . . . 925.2.1 Model Identification . . . . . . . . . . . . . . . . . . . 925.2.2 IMC design for unstable systems . . . . . . . . . . . . 935.2.3 PID implementation of IMC controller . . . . . . . . . 96
5.3 PI Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 965.4 Small-Scale Experiments . . . . . . . . . . . . . . . . . . . . . 97
5.4.1 Experimental setup . . . . . . . . . . . . . . . . . . . . 975.4.2 Experiment 1: IMC-PID controller at Z=20% . . . . . 975.4.3 Experiment 2: PI tuning at Z=20% . . . . . . . . . . 975.4.4 Experiment 3: IMC-PID controller at Z=30% . . . . . 985.4.5 Experiment 4: PI tuning at Z=30% . . . . . . . . . . 1005.4.6 Experiment 5: H∞ mixed-sensitivity controller at Z=30%1005.4.7 Experiment 6: H∞ loop-shaping controller at Z=30% 102
5.5 Medium-Scale Experiments . . . . . . . . . . . . . . . . . . . 1035.5.1 Experimental setup . . . . . . . . . . . . . . . . . . . . 1035.5.2 Experiment 7: IMC-PID controller at Z=18% . . . . . 1065.5.3 Experiment 8: PI tuning at Z=18% . . . . . . . . . . 107
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6 NONLINEAR CONTROL SOLUTIONS 1096.1 Pipeline-Riser System . . . . . . . . . . . . . . . . . . . . . . 110
6.1.1 Simplified dynamical model . . . . . . . . . . . . . . . 1106.1.2 Experimental setup . . . . . . . . . . . . . . . . . . . . 112
6.2 State Feedback with Nonlinear Observer . . . . . . . . . . . . 1126.2.1 State feedback . . . . . . . . . . . . . . . . . . . . . . 115
viii Contents
6.2.2 Nonlinear Observers . . . . . . . . . . . . . . . . . . . 1156.2.3 Experimental results . . . . . . . . . . . . . . . . . . . 1196.2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.3 Output Linearizing Controller . . . . . . . . . . . . . . . . . . 1286.3.1 Cascade system structure . . . . . . . . . . . . . . . . 1286.3.2 Stability analysis of cascade system . . . . . . . . . . 1286.3.3 Stabilizing Feedback Control . . . . . . . . . . . . . . 1316.3.4 Analogy with conventional cascade control . . . . . . . 1346.3.5 Experimental Results . . . . . . . . . . . . . . . . . . 1346.3.6 Controllability Limitations . . . . . . . . . . . . . . . 1366.3.7 Remarks on output-linearizing controller . . . . . . . . 143
6.4 PI Tuning Considering Nonlinearity . . . . . . . . . . . . . . 1436.4.1 Simple model for static nonlinearity . . . . . . . . . . 1436.4.2 PI tuning rules based on static model . . . . . . . . . 1446.4.3 Experimental result . . . . . . . . . . . . . . . . . . . 1456.4.4 Olga Simulation . . . . . . . . . . . . . . . . . . . . . 146
6.5 Gain-Scheduling IMC . . . . . . . . . . . . . . . . . . . . . . 1486.6 Comparing Four Nonlinear Controllers . . . . . . . . . . . . . 1506.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
7 CONCLUSIONS AND FUTURE WORK 1597.1 Main Contributions . . . . . . . . . . . . . . . . . . . . . . . . 159
7.1.1 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . 1597.1.2 Control structure design . . . . . . . . . . . . . . . . . 1607.1.3 PID and PI tuning . . . . . . . . . . . . . . . . . . . . 1617.1.4 Nonlinear control . . . . . . . . . . . . . . . . . . . . . 161
7.2 Future works . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
A GAS-LIFTED OIL WELL MODEL 163
B MODEL IDENTIFICATION CALCULATIONS 171
C STATIC NONLINEARITY PARAMETERS 175
D CALCULATION OF FRICTION AND DENSITY 177
REFERENCES 179
Chapter 1
INTRODUCTION
The world energy demand has been increasing rapidly due to further indus-trialisation and emerging new industrial giants such as China and India. Inspite of recent developments in new energies, fossil fuels are still the mostimportant source of energy and is expected to remain so. Because of in-creased demand, the crude oil price has stayed above 100 US$ during thepast decade. Since the resources are limited, many technologies have beendeveloped to maximize the production from existing petroleum reservoirs.Production from unconventional sources like Oil Sands in Canada, whichwas not economically feasible before, is possible now.
The first fields that were explored in the history of oil and gas were easyto produce and not very technically demanding. However, as these resourcesare drained, the demand for more advanced technology is growing. Oneinteresting fact is that even if the newly discovered offshore fields are locatedat places with deeper water, the depth of the reservoir formations tend tobe smaller. The result is that the reservoirs have low energy, meaning lowertemperatures and lower pressure and therefore making it more difficult toexploit them. The driving force for producing the oil and gas is then smallermaking the fields more prone to slugging, and the low temperatures makesit more difficult to avoid solids formation, like hydrates and waxes [63].
Safe and economical transport of the produced oil and gas by subseapipelines is a big challenge in offshore oil production. This has led to ad-vances in multiphase transport technology and the emergence of the mul-tidisciplinary field of “Flow Assurance”. In this Chapter, first, we brieflyintroduce the main processes involved in offshore oil production. Next, wediscuss challenges related to flow assurance at offshore oilfields. One ofthese challenges is severe-slugging flow which provides the main motivationfor this thesis.
1
2 INTRODUCTION
1.1 Offshore Oil Production
The main process units involved in offshore oil production are the reservoir,subsea oil wells, subsea manifolds, flowlines, risers and separators. We willbriefly introduce these processes.
1.1.1 Petroleum reservoir
A reservoir is a rock formation of sedimentary origin (with very few excep-tions) containing liquid and/or gaseous hydrocarbons. The reservoir rock isporous and permeable, and the structure is bounded by impermeable barri-ers which trap the hydrocarbons. The vertical arrangement of the fluids inthe structure is governed by gravitational forces. Figure 1.1 shows a cross-section of a typical hydrocarbon reservoir (classic anticline) [9]. Petroleumreservoirs are broadly classified as oil or gas reservoirs. These broad classi-fications are further subdivided depending on [3]:
� The composition of the reservoir hydrocarbon mixture
� Initial reservoir pressure and temperature
� Pressure and temperature of the surface production
Usually a single pressure value is referred to as the reservoir pressure or theformation pressure, and a single temperature is referred to as the reservoirtemperature. For example, the Elgin Franklin field in the North Sea hasa reservoir pressure of 1000 bar and a temperature of 200 ◦C [69]. Somereservoirs contain heavy oils that require artificial lift for production. Thecold temperatures in deepwater or in the Arctic also play a role in how wellthe oil flows. Sometimes water or produced gas is injected into the reservoirto maintain the pressure and force the oil to flow toward the productionwells.
1.1.2 Subsea oil wells
An oil well generally refers to any boring through the earth that is designedto find and acquire petroleum hydrocarbons. In particular, it is a producingwell with oil as its primary commercial product. Oil wells almost alwaysproduce some gas and frequently produce water. Most oil wells eventuallyproduce mostly gas or water [66].
Subsea wells are essentially the same as onshore wells. Mechanicallyhowever, they are placed in a Subsea structure (template) that allows thewells to be drilled and serviced remotely from the surface, and protected
1.1. Offshore Oil Production 3
Oil
Source rock
WaterPer
mea
ble
Res
ervo
ir
Rock
Source Rock
Gas
Figure 1.1: Cross-section of a typical hydrocarbon reservoir
from damage, e.g. from trawlers. The wellhead is placed in a slot in thetemplate where it mates to the outgoing pipeline as well as hydraulic andelectric control signals. Operations are handled from the surface where ahydraulic power unit (HPU) provides hydraulic power to the subsea installa-tion via an umbilical. The umbilical is a composite cable containing tensionwires, hydraulic pipes, electrical power and control and communication sig-nals [11]. The oil, water and gas sometimes travel from the reservoir tothe surface under their own pressure (natural drive). A well in which theformation pressure is sufficient to produce oil at a commercial rate withoutrequiring a pump is called a natural flowing well. Most reservoirs are ini-tially at pressures high enough to allow a well to flow naturally. If reservoirpressures are low, however, artificial lift is employed. Artificial lift can bein the form of in-well or seafloor pumps and is sometimes accompanied within-well heating and/or gas lift systems.
1.1.3 Subsea processing
A subsea processing unit (Figure 1.2) separates water from the producedmultiphase mixture and injects the water back to a disposal reservoir. Inaddition, it includes a booster to send oil and gas with a higher pressurethrough the subsea flowlines. Produced sand from the well also is treatedby the subsea processing system. Removing the water helps to improverecovery from low-pressure reservoirs by reducing the head pressure in risersand hence the back pressure on oil wells. The recovery rate of Tordis fieldof Norway has increased from 49% to 55% by use of the subsea processingsystem. Moreover, injecting the water into the reservoir reduces the wateremission from topside facilities to the sea [70].
4 INTRODUCTION
Subsea Separator
Seabed
Water
Oil & Gas
Well
Flowline
Injection
Figure 1.2: Schematic presentation of a subsea processing system
1.1.4 Subsea pipelines and risers
Pipelines (and risers) are the blood vessels of the oilfield; they are used fora number of purposes in the production of offshore hydrocarbon resources(see Figure 1.3). These include e.g. [4]:
� Export (transportation) pipelines;
� Flowlines to transfer product from platforms to export lines;
� Water injection or chemical injection flowlines;
� Flowlines to transfer product between platforms, subsea manifolds andsatellite wells;
� Pipeline bundles.
In offshore field development, pipelines are a major cost and many criteria,depending on depth and temperature must be taken into account in theirdesign. A riser system is essentially a conductor pipe connecting the well-head at the seabed to the topside facility. There are basically two kinds ofrisers; namely rigid risers and flexible risers. A hybrid riser is the combi-nation of both. The six main configurations for flexible risers are shown inFigure 1.4. Configuration design drivers include a number of factors such aswater depth, host vessel access/hang-off location, field layout such as num-ber and type of risers and mooring layout, and in particular environmentaldata and the host vessel motion characteristics [4].
1.1.5 Topside separators
The function of the topside facilities is to separate the oil well stream intothree “components” or phases (oil, gas, and water), and process these phasesinto marketable products or dispose them in an environmentally acceptable
1.1. Offshore Oil Production 5
Flowlines
Sattelite subsea
wellsSubsea
manifold
Flowlines
Riser
Export pipeline
To Shore
Figure 1.3: Subsea pipelines and riser
Free Hanging Catenary
Lazy-S
Steep Wave
Steep-S Plaint Wave
Lazy Wave
Figure 1.4: Flexible riser configurations
6 INTRODUCTION
Inlet
Gas
outlet
Water
outlet
Oil
outlet
Gas
Oil
Water
Figure 1.5: Topside three-phase separator
manner. In the separators, gas is flashed from liquids and “free water” isseparated from the oil. This should remove enough light hydrocarbons toproduce a “stable” crude oil with volatility (i.e., vapor pressure) to meetsales criteria. Separators are classified as two-phase if they separate gasfrom the total liquid stream and three-phase if they also separate the liquidstream into its crude oil and water components [65]. Figure 1.5 shows athree-phase separator. During the early years of the oil industry the gaswas burnt in flares, but today the gas is usually compressed and treated forsale or injection.
In Figure 1.5, two level controllers regulate the levels of water and oilin the separator and one pressure controllers regulate the pressure of thegas by actuating their corresponding outlet valves. Because of the couplingbetween the two levels and the pressure, the separator is an interestingmulti-variables control problem. The level switches and other safety logicsare also important requirements in operation.
1.2 Flow Assurance Challenges
“Flow assurance” is a relatively new term in the oil and gas industry. Itrefers to ensuring successful and economical flow of the hydrocarbon streamfrom the reservoir to the point of sale and is closely linked to multiphaseflow technology.
The concept of Flow Assurance developed because traditional approachesare inappropriate for deepwater production due to extreme distances, depths,temperatures or economic constraints. The term Flow Assurance was firstused by Petrobras in the early 1990s as “Garantia do Escoamento” [Por-tuguese], literally meaning “Guarantee of Flow”, or Flow Assurance [32].
Flow assurance is an extremely diverse subject matter, encompassing
1.2. Flow Assurance Challenges 7
many separate and specialized subjects and embracing all kinds of engi-neering disciplines. Besides network modeling and transient multiphasesimulation, flow assurance involves handling solid deposits, such as gas hy-drates, asphaltene, wax, scale and naphthenates (oil and condensate). Flowassurance is the most critical task during deep-water production because ofthe high pressures and low temperature involved. The financial loss fromproduction interruption or asset damage due to flow assurance mishap canbe astronomical. What compounds the flow assurance task even further isthat these solid deposits can interact with each other and cause blockage ofthe pipelines and result in flow assurance failure.
Flow Assurance is applied during all stages of system selection; detaileddesign, surveillance, troubleshooting operation problems and increased re-covery in late life etc.
1.2.1 Hydrates
Gas hydrates are crystalline materials where water molecules form a frame-work containing cavities which are occupied by individual gases or gas mix-tures (e.g. methane, ethane, propane, isobutane and inorganic moleculessuch as CO2 and H2S). Hydrates form when light hydrocarbons meet withwater, typically at temperatures less than 5◦C and at elevated pressures.
Hydrates are similar to snow in appliance and structure, but since thegaseous molecules “stabilize” the water crystalline structures, hydrates areformed before the appearance of snow. Hydrates can cause blockage ingas flowlines, and an effective way is needed to inhibit their formation. Themost common solution is use of a chemical “inhibitor” such as MEG (Mono-Ethylene Glycol) or Methanol. Another prevention measure which is usedtogether with chemicals, is insulation of pipelines. In some fields, insulationcan be used to keep the temperature above hydrate formation temperature.The calculation of thermal insulation requires thermal calculations whichcan be extensive.
If, despite of prevention strategies, hydrates are formed, it is importantto have means to remove the blockages in the system. Depressurizationis the most effective remediation mean. Direct electrical heating can be asolution in critical places. It is used in the Asgard field of Norway.
“Cold flow” is an alternative technology on hydrate prevention in pipelinesat deepwater production. This technology aims to eliminate the need forinjection of chemicals and heating under normal operating conditions atseabed. Cold flow is based on slurry transport of hydrate particles and pos-sibly other solids, like wax. However, it will only occur when we have reachedsteady-state operating conditions. The main goal for this technology is to
8 INTRODUCTION
have a solution which allows subsea field development based on long mulit-phase wellstream transport. The only problem that one may have is at thestart-up and shut-in of cold flow technology operations. These operationscan be managed by use of chemicals, or pipe insulation and heating. What-ever system will be selected for start-up and shut-in operation, it will addcost and complication to deepwater production facilities [53]. Both NTNUand SINTEF have done research work in the area cold flow technology.
1.2.2 Wax
Wax is a class of hydrocarbons that are natural constituents in any crude oiland most gas condensates. Waxy oils may create problems in oil productionfor three main reasons [32]:
� Restricted flow due to reduced inner diameter in pipelines and in-creased wall roughness
� Increased viscosity of the oil
� Settling of wax in storage tanks
For any pipeline experiencing wax deposition, there has to be a wax controlstrategy. Most often, the wax control strategy simply consists of scrapingthe wax away from the pipe wall by regular pigging. Sometimes, substantialquantities of wax are removed from the line. In one case, several tonnes ofwax were collected in the pig trap at Statfjord field in the Norwegian Sea.
1.2.3 Asphaltenes
Asphaltenes are usually the heaviest fractions of the crude oil and are definedby their solubility characteristics. They are soluble in aromatic solvents suchas toluene but precipitate upon addition of n-alkanes such as n-heptane.During production, asphaltenes are also known to precipitate as a result ofchange in pressure, temperature and or composition of the fluid.
From literature and past history, it is known that asphaltene precipita-tion is more likely to occur in an under-saturated, light reservoir fluid thana heavy hydrocarbon system. It is also noteworthy that problems due toasphaltenes occur in a two-step process: (i) precipitation from the reser-voir fluid, and (ii) deposition of the precipitated asphaltene particles. Theycause production rate decline and various other operational problems, suchas higher viscosity and water oil emulsion, etc. It is a key risk to handle inflow assurance and production chemistry.
1.2. Flow Assurance Challenges 9
Presently, The main focus of research activities is to understand asphal-tene behaviour. New techniques are being developed to measure asphaltenedeposition but are not in common use in the industry. It is a consensus inthe industry that the “prevention” approach is the best to tackle possibleproblems caused by solids deposition [46].
1.2.4 Scales
Oilfield scale is mainly deposits of inorganic salts such as carbonates and sul-phates of barium, strontium or calcium. Scale may also be salts of iron likesulphides, carbonates and hydrous oxides. Scales reduce transport capac-ity of flowlines and they can cause plugging. Scale inhibitors are chemicalswhich stop or interfere with the nucleation, precipitation and adherenceof mineral deposits. Scale dissolvers are chemicals which dissolve scale bycomplexing with ions like barium, strontium, calcium and iron. Other tech-niques like electromagnetic inhibitor exist too [79].
1.2.5 Corrosion
When “carbon steel” pipes are used in transporting oil and multiphase flowcontaining a fraction of water, there is usually a high risk of corrosion. Thedecision to use carbon steel is usually economic, in order to minimise capitalexpenditure, and its use usually requires implementing a full internal cor-rosion management strategy to control corrosion levels through the systemlife.
Various mechanisms have been postulated for the corrosion process. Allinvolve the formation of carbonic acid ion or biocarbonate when CO2 isdissolved in water.
CO2 +H2O → CO2−3 + 2H+
CO2 +H2O ↔ HCO−
3 +H+
Including iron the overall reaction is:
CO2 +H2O + Fe → FeCO3 +H2
This process can lead to corrosion of the material at a rate which is greaterthan that from general acid corrosion having the same pH value. The mecha-nism of corrosion is dependent on the quantity of CO2 dissolved in the waterphase, and predictions of corrosion levels are currently based on the knowl-edge of CO2 partial pressure and the use of correlations such as DeWaard-Milliams [45].
10 INTRODUCTION
1.2.6 Emulsions
Under a combination of low ambient temperature and high fluids viscositydue to inversion water cut conditions, stable emulsions can occur betweenthe water and oil phases. This can impair separation efficiency at the pro-cessing facility and thus cause loss in production [45].
Under these conditions there may be need to inject de-emulsifiers at thesubsea facilities and also to ensure that sufficient pressure is available tore-start the system following an unplanned shutdown.
1.2.7 Slugging flow
The combined transport of hydrocarbon liquids and gases can offer signifi-cant economic savings over the conventional, local, platform-based separa-tion facilities. The flow behavior of two-phase flow is much more complexthan that of single-phase flow. Two-phase flow is a process involving theinteraction of many variables. The gas and liquid phases normally do nottravel at the same velocity in the pipeline because of the differences indensity and viscosities. For an upward flow, the gas phase which is lessdense and less viscous tends to flow at a higher velocity than the liquidphase. On the other hand, in terrains with a downward slope the liquidflows with a higher velocity than the gas. Although the analytical solu-tions of single-phase flow are available and the accuracy of prediction isacceptable in industry, multiphase flow predictions, even when restrictedto a simple pipeline geometry, are in general quite complex [4]. Some ofthe flow patterns that occur in a horizontal pipeline for different velocitiesof flow are shown in Figure 1.6, and flow patterns in vertical pipelines arepresented in Figure 1.7.
Slug flow is one of the flow patterns occurring in multiphase pipelines asshown in Figure 1.6 and Figure 1.7. It is characterized by a series of liquidplugs (slugs) separated by a relatively large gas pockets. Slug flow can becaused by any of the following:
� Hydrodynamic slugs - this form of slugging occurs in horizontalpipelines due to differences in velocities of the different phases. Liquidbuilds up and forms slugs which are short, but with a high frequency.This type of slugging causes less problems than “severe slugging”.
� Riser slugging - this type of slugging is induced by the presence ofa vertical riser. The liquid blocks the entrance to the riser so that thegas can not enter into the riser. This is the case until the pressure ofthe upstream gas exceeds the gravitational pressure of the liquid in
1.2. Flow Assurance Challenges 11
Flow direction
Low
ve
loci
ty
Elongated bubble
flow
Dispersed
bubble flow
Annular flow
Inte
rme
dia
te v
elo
city
Hig
h v
elo
city
Slug flow
Stratified flow
Stratified wavy
flow
Figure 1.6: Flow patterns in a horizontal pipeline for different velocities
Flo
w d
ire
ctio
n
Low velocity
Ch
urn
flo
w
Dis
pe
rse
d b
ub
ble
flo
w
An
nu
lar
flo
w
Intermediate velocity High velocity
Slu
g f
low
Figure 1.7: Flow patterns in a vertical pipeline for different velocities
12 INTRODUCTION
the riser column. This type of slugging causes long liquid slugs andlarge pressure variations. It is also known as “severe slugging”.
� Terrain-induced slugging - induced by irregular surface of theseabed, liquid tends to accumulate at places with lower elevations.This can cause blockage of the gas flow in the pipe.
� Operational-based slugging - one example is pig-induced sluggingthat happens when a pig is running for cleaning the wax.
� Casing heading slugs in gas-lifted wells - this type of slugging issimilar to the riser slugging, but gas is compressed in the annulus ofthe well.
� Density-wave slugs in long risers and wells - accumulation ofgas at the bottom of the riser (well) makes a region with low-densityand this region travels upward.
Slugging in production pipelines and risers has for many years been a majoroperational problem in subsea oil-gas fields developments. The slugging canresult in severe fluctuations in pressure and flow rate (gas and liquid) at boththe wells end and the receiving host processing facilities. This causes manyproblems, e.g. [43]:
� Oscillations are not in agreement with smooth operation
� Safety aspects and shutdown risks
� The total oil and gas production must usually be less than the systemsdesign capacity to allow for the peak production
� Unstable mode often decreases sharply the lift gas efficiency for gas-lifted wells
� Difficulties with gas-lift allocation computation due to instabilities
� Drawbacks on facilities, well operations and equipments
1.3 Riser Slugging
The riser-slugging instability is one of the main flow assurance challengesdiscussed in the previous section. In the following, we describe the mecha-nism of the riser slugging in details. Next, we summarise the conventionalsolutions to prevent the riser slugging. Finally, automatic control is intro-duced as an anti-slug solution with a large potential of economical benefit.
1.3. Riser Slugging 13
(1) Slug Formation (2) Slug Production
(3) Blowout(4) Liquid Fallback
Figure 1.8: Schematic of severe slugging in flowline-riser systems
1.3.1 Mechanism of riser slugging
A downward inclination of the pipeline ending to the riser increases thepossibility of the riser slugging. This enables liquid to accumulate in theentrance to the riser, and causes liquid to block the entrance in to the riser.This leads to compression of the gas in the pipeline and finally expansion ofthe gas in the riser when the gas pressure is higher than the hydrostatic headin the riser [44]. The mechanism of riser slug formation can be describedby the following four steps:
� Step 1 - Liquid accumulates in the riser low-point due to gravity.This is the case if the gas and liquid velocities are low enough to allowfor it.
� Step 2 - As long as the hydrostatic head of the liquid in the riser ishigher than the pressure drop over the riser the slug continues to growas gas can not penetrate the liquid blocking the entrance.
� Step 3 -When the pressure drop over the riser exceeds the hydrostatichead, the liquid is pushed out of the riser.
� Step 4 - When all the liquid has left the riser the velocities are sosmall that liquid falls back in to the low-point of the riser and startsto accumulate again.
The four steps are illustrated in Figure 1.8 (adapted from [63]).
The flow pattern in risers is not always severe slugging. However, itdepends on inflow conditions, topside choke valve, geometry and dimensionsof the riser as well as the separator pressure. Flow regime maps are used to
14 INTRODUCTION
7.04 2.83 2.63 1.41
13.96
4.01
2.70
6.43
Figure 1.9: Experimental setup for S-riser at NTNU
study the stability of the flow in risers where the flow regime is shown fordifferent values of gas superficial velocity Usg and liquid superficial velocityUsl. One example of such maps is shown in Figure 1.10. This flow regimemap was obtained from experiments on a setup for S-riser at multiphaseflow laboratory of NTNU [59]. The Geometry and dimensions of the riserare given in Figure 1.9 and the separator is at the atmospheric pressure.Two types of severe-slugging flow are observed at low velocities of liquidand gas inflows. Severe slugging I gives slugs with larger amplitudes (about1.8 bar peak of inlet pressure) compared to those of severe slugging II (about1.4 bar peak of the inlet pressure). By increasing the inflow velocities theflow regime changes to transient slugging with small but high frequencyslugs. The flow becomes stable for sufficiently large inflow rates which canbe dispersed bubble or annular flow regimes.
1.3.2 Conventional anti-slug solutions
Slug catcher: A slug catcher is a vessel with sufficient buffer volume tostore the largest slugs expected from the upstream system. The slug catcheris located between the outlet of the riser and the processing facility. Thebuffered liquids can be drained to the processing equipment at a muchslower rate to prevent overloading the system. As slugging is a periodicalphenomenon, the slug catcher should be emptied before the next slug ar-rives. To design the slug catcher many criteria depending of flow rates andlength of the slugs are considered. This solution increases the capital costand the slug catcher tank occupies space on the platform.
1.3. Riser Slugging 15
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
0 1 2 3 4 5
Usl [m
/s]
Usg [m/s]
SS1
SS2
TTS
STABLE
Severe Slugging I
Severe Slugging II
Transient Slugging
Stable Flow
Figure 1.10: Flow regimes occurring in S-riser
Topside Choking: Closing the topside choke valve can eliminate severe-slugging [68]. However, this increases back pressure of the subsea oil wellsand decreases the production rate. Topside chocking was one of the firstmethods proposed to prevent the severe slugging phenomenon in 1973 [82].It was observed that increased back pressure could eliminate severe sluggingbut would reduce the flow capacity severely.
Riser Base Gas Injection: This method provides artificial lift for theliquids, moving them steadily through the riser. This technique can alleviatethe problem of severe slugging by changing the flow regime from slug flowto annular or dispersed flow, but does not help with transient slugging. Itis one of the most frequently used methods for current applications [57].
The riser base gas injection method was first used to control hydrody-namic slugging in vertical risers. However, it was dismissed as not beingeconomically feasible due to the cost of a compressor for pressurizing thegas for injection and the piping required to transport the gas to the base ofthe riser [67]. Loss and Joule–Thomson cooling are the potential problemsresulting from high-injection gas flow rates.
To reduce amount of the injected gas, automatic feedback control canbe used. Control of the hold-up at top of the riser by manipulating the gasinjection valve has been proposed in [52].
Full Separation: Another solutions is “full separation” where liquid
16 INTRODUCTION
Inlet
separator
Subsea
wells
Riser
Topside
chokePT
zPset PC
Prb
Figure 1.11: Preventing slug flow by control of riser base pressure
phases and gas are separated at the subsea and transported to the topsidefacilities by separate risers. However, this method is not economical.
1.3.3 Automatic control solution
Most of the researchers in fluid dynamics believe that the boundaries be-tween flow regimes shown in Figure 1.10 are given by nature and cannot bemoved without changing riser geometry or the boundary conditions. How-ever, automatic feedback control can change the dynamics of a given steady-state operating point from unstable to stable. By use of automatic control,it is possible to stabilize a normally unstable “non-slugging” flow regime atconditions that the system would give the “slugging” flow regime withoutcontrol. In other words, we are able to change the boundaries between theflow regimes.
The first successful use of automatic feedback control to prevent slug flowwas reported in 1996 [10]. The French company Total used anti-slug con-trol to prevent severe slugging in the Dunbar 16 inches multiphase pipelinelocated in the North Sea. They controlled the riser base pressure by manip-ulating the topside choke valve as shown in Figure 1.11. In 2000, ABB re-ported a similar use of automatic control to prevent slugging in the pipelinebetween Hod and Valhall platforms in the North Sea [30]. They used a com-bination of feedforward control and dynamic feedback control in a cascadedstructure.
Figure 1.12 shows a bifurcation diagram of the experimental S-riser pre-sented in Section 1.3.1. The bifurcation diagram illustrates the behavior ofthe system over the whole working range of the choke valve [77]. The solidlines show the maximum and minimum of the slugging oscillation and theaverage value. The dashed line shows the unstable non-slug steady-state
1.4. Motivation 17
20 30 40 50 60 70 80 90 1001
1.2
1.4
1.6
1.8
2
2.2
Pressure at inlet of pipeline (Pin
)
Pin
[bar
]
valve opening [%]
maximumaveragesteady−stateminmum
Figure 1.12: Bifurcation diagram of the experimental setup for S-riser [59]
that becomes slugging flow without control. If we manually close the chokevalve sufficiently the inlet pressure increases and the flow becomes stable.For the present experimental setup, the critical value of the relative valveopening for the transition between a stable non-oscillatory flow regime andriser slugging is Z = 26%.
The unstable steady-state is like equilibrium of an inverted pendulumand it can be stabilized by dynamic feedback control. The steady-statepressure is lower than the average pressure and the system can operated onlower pressure set-point by use of anti-slug control. This leads to a higherproduction rate from subsea oil wells and economical benefits [60].
1.4 Motivation
Existing anti-slug control systems are often not operating in practice be-cause of robustness problems; the closed-loop system becomes unstable af-ter some time, for example because of inflow disturbances or plant changes.The operators turn off the controller and instead use manual choking whenthe control system becomes unstable. The main objective of our research isto find robust solutions for anti-slug control systems. The nonlinearity atdifferent operating conditions is one source of plant change, because gain ofthe system changes drastically for different operating conditions. In addi-tion, the time delay is another problematic factor for stabilization. A robustcontroller must have a good gain margin in presence of plant changes and agood delay margin (phase margin) for transportation delay in long flowlines.Furthermore, the system should not drift away from the design operatingpoint (setpoint tracking or performance).
First, we aim at finding a good control structure, that is finding con-trolled variables, manipulated variables and parings that give a robust
18 INTRODUCTION
closed-loop system for stabilizing control. Next, we consider both linearand nonlinear approaches for the controller design. PID and PI controllersare commonly used in industry and we seek tuning rules based on closed-loop step test for robust anti-slug control. Then, we consider nonlinearmodel-based control design to counteract nonlinearity of the system. Wecompare both the linear and the nonlinear approaches and find advantagesand disadvantages of each.
1.5 Organization of Thesis
Chapter 1 (this chapter) gives an introduction to offshore oil production andflow assurance challenges associated with multiphase transport at offshoreoil fields. Then, the slugging flow is described in more details and solutionsto prevent slugging flow are reviewed.
Chapter 2 presents simplified dynamical models for severe slugging. Anew simplified model for pipeline-riser systems is presented, and this modelis extended to well-pipeline-riser system. The suggested models are com-pared to results from OLGA simulator and experiments.
Chapter 3 deals with selection of suitable controlled variables for sta-bilizing control by use of controllability analysis of the system. Controlstructures to prevent slugging flow in well-pipeline-riser system and gas liftoil wells are suggested.
In Chapter 4, we consider three different manipulated variables (controlvalves) for the well-pipeline-riser system. We choose the suitable manipu-lated variable for anti-slug control by use of controllability analysis. Then,we validate the result by OLGA simulations and experiments.
Chapter 5 proposes a new closed-loop model identification and InternalModel Control (IMC) design for anti-slug control. Furthermore, the PIDand PI tunings are obtained from the IMC controller. The proposed modelidentification and tuning rules are tested experimentally.
In Chapter 6, nonlinear model-based control designs are considered.First, we use state estimation by three different observers and state-feedback.Then, the feedback linearization design by measured outputs is applied.Next, a gain scheduling of three IMC controllers is compared to the twononlinear model-based approaches.
1.6. List of Contributions 19
In Chapter 7, the main conclusions and remarks of this thesis are sum-marized along with some suggestions for further work.
1.6 List of Contributions
1.6.1 Conference publications
1. E. Jahanshahi, S. Skogestad, Closed-loop model identification andPID/PI tuning for robust anti-slug control In: 10th IFAC Interna-tional Symposium on Dynamics and Control of Process Systems (DY-COPS). Mumbai, India, December 2013.
2. E. Jahanshahi, S. Skogestad, Comparison between nonlinear model-based controllers and gain-scheduling Internal Model Control based onidentified model. In: 52nd IEEE Conference on Decision and Control(CDC). Florence, Italy, December 2013.
3. E. Jahanshahi, S. Skogestad, E. I. Grøtli. Nonlinear model-based con-trol of two-phase flow in risers by feedback linearization. In: 9th IFACSymposium on Nonlinear Control Systems (NOLCOS). Toulouse, France,September 2013.
4. E. Jahanshahi, S. Skogestad, M. Lieungh. Subsea solution for anti-Slug Control of multiphase risers. In: European Control Conference(ECC). Zurich, Switzerland, July 2013.
5. E. Jahanshahi, S. Skogestad, E. I. Grøtli. Anti-slug control experi-ments using nonlinear Observer. In: American Control Conference(ACC). Washington D.C., USA, June 2013.
6. E. Jahanshahi, S. Skogestad, A. H. Helgesen. Controllability analysisof severe slugging in well-pipeline-riser systems. in: IFAC Workshop -Automatic Control in Offshore Oil and Gas Production. Trondheim,Norway, May 2012.
7. E. Jahanshahi, S. Skogestad, H. Hansen. Control structure design forstabilizing unstable gas-lift oil wells. in: International Symposium onAdvanced Control of Chemical Processes (ADCHEM), Singapore, July2012
8. E. Jahanshahi and S. Skogestad, Simplified dynamical models for con-trol of severe slugging in multiphaser risers. in: Proceeding of the
20 INTRODUCTION
18th International Federation of Automatic Control (IFAC) WorldCongress 1634–1639. Milan, Italy, September 2011.
1.6.2 Journal publications
1. E. Jahanshahi, S. Skogestad, Simplified dynamical models for controlof severe slugging flow in offshore oil production, In Preparation.
2. E. Jahanshahi, S. Skogestad, Control structure design for stabilizingcontrol: anti-slug control at offshore oilfields, In Preparation.
3. E. Jahanshahi, S. Skogestad, Linear control solutions for anti-slugcontrol at offshore oilfields, In Preparation.
4. E. Jahanshahi, S. Skogestad, Nonlinear control solutions for anti-slugcontrol at offshore oilfields, In Preparation.
1.6.3 Patents
1. Title: Pipeline-riser system and method of operating the samePatent No. 13174514.3 - 1605‘Subsea solution for anti-slug control at offshore oilfields’Inventors: Sigurd Skogestad, Esmaeil Jahanshahi
2. Title: Method of operating a pipeline-riser systemPatent No. 13174513.5 - 1605‘Closed-loop model identification and PID/PI tuning for robust anti-slug control’Inventors: Esmaeil Jahanshahi, Sigurd Skogestad
3. Title: Robust PID tuning based on relay-feedback and IMC design‘Robust PID tuning based on relay-feedback and IMC design’Inventors: Esmaeil Jahanshahi, Selvanathan Sivalingam, Brad Schofield
Chapter 2
SIMPLIFIED DYNAMICAL
MODELS
Instead of elaborated models such as those used for simulation purposes(e.g. OLGA simulator), a simple dynamical model with few state variablesis desired for model-based control design. We propose a new simplified dy-namic model for severe slugging flow in pipeline-riser systems. The proposedmodel, together with five other simplified models found in the literature, arecompared with results from the OLGA simulator. The new model can beextended to other cases, and we consider also a well-pipeline-riser system.The proposed simple models are able to represent the main dynamics ofsevere slugging flow and compare very well with OLGA simulations andexperiments.
2.1 Introduction
Slugging has been recognised as a serious problem in offshore oilfields andmany efforts have been made in order to prevent this problem [10], [30].The severe slugging flow regime usually occurs in pipeline-riser systemsthat transport oil and gas mixture from the seabed to the surface. Thisproblem, also referred to as “riser slugging”, is characterised by severe flowand pressure oscillations. Slugging problems have also been observed in gas-lifted oil wells where two types of instabilities, casing heading and densitywave instability, have been reported [31].
The irregular flow caused by slugging can cause serious operational prob-lems for the downstream surface facilities, and an effective way to handleor remove riser slugging is needed. The conventional solution is to reducethe opening of the top-side choke valve (choking), but this may reduce the
21
22 SIMPLIFIED DYNAMICAL MODELS
production rate especially for fields where the reservoir pressure is rela-tively low. Therefore, a solution that guarantees stable flow together withthe maximum possible production rate is desirable. Fortunately, automaticcontrol has been shown to be an effective strategy to eliminate the sluggingproblem and different slug control strategies have tested experimentally [25].Anti-slug control is an automatic feedback control system that aims at sta-bilising the flow in the pipeline at the same operating conditions that un-controlled would yield riser slugging [75]. This control system usually usesthe top-side choke as the manipulated variable and the riser base pressuresas the controlled variable.
There have been some research on riser slugging using the OLGA simu-lator to test anti-slug control [21], but for controllability analysis and con-troller design a simpler dynamical model of the system is desired. The focusof this chapter is on deriving the simple dynamical models which capturethe essential behaviour for control. For control, it is more important tocapture the main dynamics for the onset of slugging, not the slugging itself.The aim is to avoid the slug flow regime and instead operate at a steady(non-slug) flow regime. Therefore, The shape and length of the slugs arenot main concerns in this modeling.
Five simplified dynamical models for the pipeline-riser systems werefound in the literature. The “Storkaas model” [76] is a three-dimensionalstate-space model which has been used for controllability analysis [77]. The“Eikrem model” is four dimensional state-space model [80], [17]. Anothersimplified model, referred to as the “Kaasa model” [49], only predicts thepressure at the bottom of the riser. The “Nydal model” [55] is the onlymodel that includes friction in the pipes. The most recently published sim-plified model is the “Di Meglio model” [13], [14]. In addition, we present anew four-state model which includes useful features of the other five models.The six models are simulated in the time domain and compared to resultsfrom the more detailed OLGA model in the following five aspects, listed inorder of importance:
� Critical valve opening for onset of slugging
� Frequency of oscillations at the critical point (onset of slugging)
� Dynamical response to a step change in the valve opening (non-slugregime)
� Steady-state pressure and flow rate values (non-slug regime)
� Maximum and minimum (pressure and flow rate) of the oscillations(slug regime)
2.2. New simplified four-state model 23
The simplified models are also analysed linearly in the frequency domainwhere we consider the location of unstable poles and important unstable(RHP) zeros in the model. The results presented in this chapter have beenpartially presented by [34], and [40]. In the present chapter, we comparethe new model to the experiments and we extend it to a well-pipeline-risersystem. This chapter is organised as follows. In Section 2, we presentour new simplified model for pipeline-riser systems. Then, In Section 3,we compare the proposed model and the other simple models to the resultsfrom the OLGA simulator and experiments. Finally, in Section 4, we extendthis model to well-pipeline-riser systems and compare the extended modelto OLGA simulations.
2.2 New simplified four-state model
2.2.1 Mass balance equations for pipeline and riser
For the new simplified model, consider the schematic presentation of thesystem in Fig. 2.1. The four differential equations in the proposed modelare simply the mass conservation law for the gas and liquid phases in thepipeline and riser sections:
dmg,p
dt= wg,in − wg,rb (2.1a)
dml,p
dt= wl,in − wl,rb (2.1b)
dmg,r
dt= wg,rb − wg,out (2.1c)
dml,r
dt= wl,rb − wl,out (2.1d)
The four state variables in the model are
� mg,p: mass of gas in pipeline [kg]
� ml,p: mass of liquid in pipeline [kg]
� mg,r: mass of gas in riser [kg]
� ml,r: mass of liquid in riser [kg]
The model is described in detail below and its main parameters are given inTable 2.1. Four tuning parameters, further described below, can be used tofit the model to the experimental or numerical data for given pipeline-risersystem.
24 SIMPLIFIED DYNAMICAL MODELS
a) Simplified representation of desired flow regime
b) Simplified representation of liquid blocking leading to riser slugging
wg,in
wl,in
wout
Ps
h
Lr
hd
mg,p Pin Vg,p !g,p "l,p
mg,r Prt Vg,r !g,r "l,rt
Choke valve with opening Z1
ml,r
h<hd
wg,rb
wl,rb
ml,p
Lp
Lh
wg,in
wl,in
Lh
mg,p Pin Vg,p g,p !l,p
Ps
Choke valve with opening Z1
h
Lr
hd
mg,r Prt Vgr !g,r "l,rt
ml,rh>hd
wg,rb= 0
wl,rb
ml,p
Lp
wout
Figure 2.1: Pipeline-riser system with important parameters
� Kh: correction factor for level of liquid in pipeline
� Cv1: production choke valve constant
� Kg: coefficient for gas flow through low point
� Kl: coefficient for liquid flow through low point
2.2.2 Inflow conditions
In equations (2.1a) and (2.1b), wg,in and wl,in are the inlet gas and liq-uid mass flow rates. They are here assumed to be constant, but the inlet
2.2. New simplified four-state model 25
boundary conditions can easily be changed, for example to make the inletflow pressure-driven.
2.2.3 Outflow conditions
We consider a constant pressure (separator pressure, Ps) as the outletboundary condition and a simple choke valve equation determines the out-flow of the two-phase mixture.
wout = Cv1f(z1)√
ρrtmax(Prt − Ps, 0), (2.2)
Here 0 < z1 < 1 is the normalized valve opening (we use the ‘capital’ Z1
when the valve opening is given in percentage, 0 < Z1 < 100) and f(z1) isthe characteristic equation of the valve. In the simulations, a linear valve isused, i.e. f(z1) = z1, but this should be changed for other valves. wl,out andwg,out, the individual outlet mass flow rates of liquid and gas, are calculatedas follows,
wl,out = αml,rtwout, (2.3)
wg,out = (1− αml,rt)wout. (2.4)
Here, αml,rt, the liquid mass fraction at top of the riser, is given by
αml,rt =
αl,rtρlαl,rtρl + (1− αl,rt)ρg,r
. (2.5)
The density of the two-phase mixture at top of the riser in (2.2) is
ρrt = αl,rtρl + (1− αl,rt)ρg,r. (2.6)
The liquid volume fraction, αl,rt in (2.5) and (2.6), is calculated by equation(2.42).
2.2.4 Pipeline model
We now introduce some important parameters. The liquid volume fraction,αl, in the pipeline section is given by the liquid mass fraction, αm
l , anddensities of the two phases [7]:
αl =αml /ρl
αml /ρl + (1− αm
l )/ρg
The average liquid mass fraction in the pipeline section is assumed to begiven by the inflow boundary condition:
αml,p =
wl,in
wg,in + wl,in
26 SIMPLIFIED DYNAMICAL MODELS
The average liquid volume fraction in the pipeline is then
αl,p =ρg,pwl,in
ρg,pwl,in + ρlwg,in. (2.7)
The gas density ρgp is calculated based on the nominal pressure (steady-state) in the pipeline by assuming ideal gas,
ρg,p =Pin,nomMg
RTp(2.8)
Here, Pin,nom, which depends itself on αl,p, is calculated from a steady-stateinitialization of the overall model. By using (2.8) and constant (nominal)inflow rates, we get αl,p in (2.7) as a constant parameter.
The cross section area of the pipeline is
Ap =π
4D2
p, (2.9)
where Dp is the diameter of the pipeline, then the volume of the pipeline isVp = ApLp. When gas and liquid are distributed homogeneously along thepipeline, the mass of liquid in the pipeline is
ml,p = ρlVpαl,p. (2.10)
With this assumption, the level of liquid in the pipeline at the low-point isgiven approximately by h ≈ hdαl,p where hd = Dp/cos(θ) is the pipelineopening at the riser base and θ is the inclination of the pipeline at thelow-point. More precisely, we use in the model
h = Khhdαl,p (2.11)
where Kh is a correction factor around unity which can be used for fine-tuning the model. If the liquid content of the pipeline increases by ∆ml,p, itstarts to fill up the pipeline from the low-point. A length of pipeline equalto ∆L will be occupied by only liquid, where
∆ml,p = ml,p − ml,p = ∆LAp(1− αl,p)ρl
and the level of liquid in the pipeline becomes h = h+∆L sin(θ) or
h = h+
(
ml,p − ml,p
Ap(1− αl,p)ρl
)
sin(θ). (2.12)
Thus, the level of liquid in the pipeline, h, can be written as a function ofliquid mass in the pipeline ml,p which is a state variable of the model. Therest of the parameters in (2.12) are constants.
2.2. New simplified four-state model 27
The pipeline gas density is
ρg,p =mg,p
Vg,p, (2.13)
where the volume occupied by gas in the pipeline is
Vg,p = Vp −ml,p/ρl. (2.14)
The pressure at the inlet of the pipeline, assuming ideal gas, is
Pin =ρg,pRTp
Mg. (2.15)
We consider only the liquid phase when calculating the friction pressure lossin the pipeline [7].
∆Pfp =αl,pλpρlU
2sl,inLp
2Dp(2.16)
Here, λp is the friction factor of the pipeline which an be computed from anexplicit approximation of the implicit Colebrook-White equation [26]:
1√
λp
= −1.8 log10
[
(
ǫ/Dp
3.7
)1.11
+6.9
Rep
]
(2.17)
Here, the Reynolds number is
Rep =ρlUsl,inDp
µ(2.18)
and µ is the viscosity of liquid and Usl,in is the superficial velocity of liquid:
Usl,in =4wl,in
πDp2ρl
(2.19)
2.2.5 Riser model
Total volume of riser:Vr = Ar(Lr + Lh), (2.20)
whereAr =
π
4D2
r (2.21)
Volume occupied by gas in riser:
Vg,r = Vr −ml,r/ρl (2.22)
28 SIMPLIFIED DYNAMICAL MODELS
Density of gas in riser:
ρg,r =mg,r
Vg,r(2.23)
Pressure at top of riser from ideal gas law:
Prt =ρg,rRTr
Mg(2.24)
Average liquid volume fraction in riser:
αl,r =ml,r
Vrρl(2.25)
Average density of mixture inside riser:
ρm,r =mg,r +ml,r
Vr(2.26)
Friction loss in riser:
∆Pfr =αl,rλrρm,rU
2m(Lr + Lh)
2Dr(2.27)
Friction factor of riser using same correlation as for pipeline:
1√λr
= −1.8 log10
[
(
ǫ/Dr
3.7
)1.11
+6.9
Rer
]
(2.28)
Reynolds number of flow in riser:
Rer =ρm,rUmDr
µ(2.29)
Average mixture velocity in riser:
Um = Usl,r + Usg,r (2.30)
Usl,r =wl,in
ρlAr(2.31)
Usg,r =wg,in
ρg,rAr(2.32)
2.2. New simplified four-state model 29
2.2.6 Gas flow model at riser base
As illustrated in Fig. 2.1(b), when the liquid level in the pipeline sectionexceeds the openings of the pipeline at the riser base (h > hd), the liquidblocks the low-point and the gas flow rate wg,rb at the riser base is zero,
wg,rb = 0, h ≥ hd (2.33)
When the liquid is not blocking at the low-point (h < hd in Fig. 2.1(a)),the gas will flow from the volume Vg,p to Vg,r with a mass rate wg,rb [kg/s]which is assumed to be given by an“orifice equation” (e.g. [73]):
wg,rb = KgAg
√
ρg,p∆Pg, h < hd (2.34)
where∆Pg = Pin −∆Pfp − Prt − ρm,rgLr −∆Pfr (2.35)
The free area Ag for gas flow can be calculated precisely using trigonometricfunctions [76], but for simplicity, a quadratic approximation is used in thenew model,
Ag = Ap
(
hd − h
hd
)2
, h < hd (2.36a)
Ag = 0, h ≥ hd (2.36b)
2.2.7 Liquid flow model at riser base
The liquid mass flow rate at the riser base is also described by an orificeequation:
wl,rb = KlAl
√
ρl∆Pl, (2.37)
where∆Pl = Pin −∆Pfp + ρlgh− Prt − ρm,rgLr −∆Pfr (2.38)
andAl = Ap −Ag (2.39)
2.2.8 Phase distribution model at outlet choke valve
In order to calculate the mass flow rates of the individual phases as givenin equations (2.2)-(2.4), the phase distribution at top of the riser must beknown. The liquid volume fraction at top of the riser, αlt, which was usedin (2.5) and (2.6), can be calculated by the entrainment model proposedby Storkaas [76], but their entrainment equations are complicated. Instead,we use the fact that in a vertical gravity-dominant two-phase pipeline there
30 SIMPLIFIED DYNAMICAL MODELS
is approximately a linear relationship between the pressure and the liquidvolume fraction. This has been observed in OLGA simulations. In addition,the pressure gradient is assumed constant along the riser for the desirednon-slugging flow regimes, which then gives that the liquid volume fraction
gradient is constant, i.e.∂αl,r
∂y = constant. It then follows that the averageliquid volume fraction in the riser is
αl,r =αl,rb + αl,rt
2(2.40)
Here, αl,r is also given by (2.25) and αl,rb is determined by the flow area ofthe liquid phase at the riser base (low-point):
αl,rb =Al
Ap(2.41)
Therefore, the liquid volume fraction at the top of the riser becomes
αl,rt = 2αl,r − αl,rb =2ml,r
Vrρl− Al
Ap(2.42)
2.3 Comparison of models
2.3.1 OLGA test case and reference model
In order to study the dominant dynamic behavior of a typical, yet simpleriser slugging problem, we use the test case for severe slugging in the OLGAsimulator. OLGA is a commercial multiphase simulator widely used in theoil industry [5]. The geometry of the system is given in Fig. 2.2. Thepipeline diameter is 0.12 m and its length is 4300 m. Starting from theinlet, the first 2000 m of the pipeline is horizontal and the remaining 2300m is inclined downwards with a 1◦ angle. This gives a 40.14 m descent andcreates a low pint at the end of the pipeline. The riser is a vertical 300 mpipe with a diameter of 0.1 m. A 100 m horizontal section with the samediameter as that of the riser connects the riser to the outlet choke valve.The feed into the system is nominally constant at 9 kg/s, with wl,in = 8.64kg/s (oil) and wg,in = 0.36 kg/s (gas). The separator pressure (Ps) after thechoke valve, is nominally constant at 50.1 bar. This leaves the choke valveopening Z1 as the only control degree of freedom (manipulated variable) inthe system.
For the present case study, the critical value of the relative valve openingfor the transition between a stable non-oscillatory flow regime and riserslugging is Z∗
1 = 5%. This is illustrated by the OLGA simulations in Fig. 2.3which show the inlet pressure, topside pressure and outlet flow rate, withthe valve openings of 4% (no slug), 5% (transient) and 6% (riser slugging).
2.3. Comparison of models 31
Figure 2.2: Geometry of OLGA pipeline-riser test case
0 50 100 15070
72
74
76
78
Z1= 4 [%]
Pin
[bar
]
0 50 100 15070
72
74
76
78
Z1= 5 [%]
0 50 100 15070
72
74
76
78
Z1= 6 [%]
0 50 100 150
52
54
56
58
60
Prt [b
ar]
0 50 100 150
52
54
56
58
60
0 50 100 150
52
54
56
58
60
0 50 100 150
5
10
15
wo
ut [k
g/s]
t [min]0 50 100 150
5
10
15
t [min]0 50 100 150
5
10
15
t [min]
Figure 2.3: Simulations of OLGA test case for different valve openings
32 SIMPLIFIED DYNAMICAL MODELS
2.3.2 Comparison of models with OLGA simulations
The different simplified models were simulated in Matlab and their tuningparameters were adjusted to match the OLGA reference model simulations.This was mostly done by trial and error, but we believe that the obtainedtuning parameters are reasonable for all the models. A more systematicapproach has been proposed in [14] for tuning the Di Meglio model, butthis approach did not work well for the present case study. The results aresummarized in Table 2.2 which shows the error (in %) for various param-eters. Our most important criteria for the model fitting are critical valueof the valve opening (Z∗
1 ) and the oscillation frequency at this point, whichfor the present case study, should be 5% and 15.6 [min].
Frequency of oscillations
All models were linearised at the critical operating point Z∗
1 = 5%. Theperiod of oscillations at this operating point is related to poles of the linearmodels. Most of the models give a pair of complex conjugate poles, s =±ωci = ±0.0067i. Note that
ωc=2π
Tc,
2π
15.6[min] 60[s/min]= 0.0067s−1
The exceptions are the Eikrem model and the Nydal model that are notable to get the right period time (15.6 [min]) for the critical valve opening,and consequently they result in different poles at Z∗
1 = 5%.
Step response
Fig. 2.4 shows the pressure response at the top of the riser to a step changein the valve opening from Z1 = 4% to Z1 = 4.2% for the OLGA referencemodel and the new model. Step responses of the OLGA model has oneundershoot and one overshoot. The amplitudes of the overshoot and un-dershoot for different simplified models are given in Table 2.2 in the formof errors from those of the OLGA model. The inverse response (overshoot)is corresponding to the RHP zeros near the imaginary axis which are alsogiven for Z1 = 5% in Table 2.2.
Bifurcation diagrams
Fig. 2.5 shows the steady-sate behaviour of the new model (central line) andalso the minimum and maximum of the oscillations compared to those of
2.3. Comparison of models 33
0 2 4 6 8 10
−1
−0.8
−0.6
−0.4
−0.2
0
0.2
time [min]
Prt [b
ar]
OLGA reference modelNew low−order model
Figure 2.4: Step response of pressure at top of riser
the OLGA model. In order to have a quantitative comparison, deviationsof the different simplified models from the OLGA reference model for fullyopen valve (Z1 = 100%) are summarised in Table 2.2.
Comparison summary
As seen in Table 2.2, there is a trade-off between model complexity andthe number of tuning parameters used to match the actual process data.Very simple models, like the Kaasa model (with seven parameters) and theDi Meglio model (with five parameters), require many parameters to geta good fit. However, finding the parameter values is difficult. The Nydalmodel and the Eikrem model (with three parameters) are also simple, butthey are not able to match the OLGA simulations because of few tuningparameters.
As opposed to the other simplified models, the new model does notrequire adjusting any physical property of the system, such as volume ofgas in the pipeline. The new model (with four parameters) is somewhatcomplicated, but is able to give a good match with relatively few tuningparameters.
The new model and the De Meglio model have approximately the sameaccuracy in prediction of the steady-state and also minimum and maximumof slugging pressures and the flow rate, but dynamically the new model iscloser to the OLGA simulations. This can be seen by comparing the stepresponse of the top-side pressure.
34 SIMPLIFIED DYNAMICAL MODELS
0 20 40 60 80 10060
65
70
75
80
85
valve opening, Z1 [%]
Pin
[bar
]
a) Pressure at inlet of pipeline (Pin
)
OLGA reference modelNew low−order model
0 20 40 60 80 10050
52
54
56
58
60
valve opening, Z1 [%]
Prt [b
ar]
b) Pressure at top of riser (Prt)
OLGA reference modelNew low−order model
0 20 40 60 80 1000
10
20
30
40
50
valve opening, Z1 [%]
wo
ut [k
g/s]
c) Outlet mass flow (wout
)
OLGA reference modelNew low−order model
Figure 2.5: Bifurcation diagrams of simplified pipeline-riser model (solidlines) compared OLGA reference model (dashed)
2.3. Comparison of models 35
Pump
Buffer
Tank
Water
Reservoir
Seperator
Air to atm.
Mixing Point
safety valve
P1
Pipeline
Riser
Top-side
Valve
Water Recycle
FT water
FT air
P3
P4
P2
Figure 2.6: Experimental setup
2.3.3 Comparison with experiments
The experiments were performed on a laboratory setup for anti-slug con-trol at the Chemical Engineering Department of NTNU. Fig. 2.6 shows aschematic presentation of the laboratory setup. The pipeline and the riserare made from flexible pipes with 2 cm inner diameter. The length of thepipeline is 4 m, and it is inclined with a 15◦ angle. The height of the riseris 3 m. A buffer tank is used to simulate the effect of a long pipe with thesame volume, such that the total resulting length of pipe would be about70 m. Other parameters and constants are given in Table 2.3.
The topside choke valve is used as the input for control. The separatorpressure after the topside choke valve is nominally constant at atmosphericpressure. The feed into the pipeline is assumed to be at constant flowrates, 4 litre/min of water and 4.5 litre/min of air. With these boundaryconditions, the critical valve opening where the system switches from stable(non-slug) to oscillatory (slug) flow is at Z∗
1 = 15%.
In addition, we developed a new OLGA case with the same dimensionsand boundary conditions as the experimental set-up. The bifurcation dia-grams are shown in Fig. 2.7 where simplified model (thin solid lines) is com-pared to the experiments (bold solid lines) and the OLGA model (dashedlines). In Fig. 2.7, the system has a stable (non-slug) flow when the top-side valve opening Z1 is smaller than 15%, and it switches to slugging flowconditions for Z1 > 15%.
36 SIMPLIFIED DYNAMICAL MODELS
0 10 20 30 40 50 60 70 80 90 10010
20
30
40
50
Pressure at inlet of pipeline, Pin
Pin
[kpa
]
topside valve opening, Z1 [%]
experiment new low−order model Olga model
0 10 20 30 40 50 60 70 80 90 1000
5
10
15
20
25
30
Pressure at top of riser, Prt
Prt [k
pa]
topside valve opening, Z1 [%]
experiment new low−order model Olga model
Figure 2.7: Bifurcation diagrams of simplified pipeline-riser model (thinsolid lines) compared to experiments (thick solid lines) andOLGA reference model (dashed lines)
2.4. Well-Pipeline-Riser System 37
2.4 Well-Pipeline-Riser System
In the pipeline-riser model described above constant gas and liquid flowrate were used as inlet boundary conditions. In order to study effect the ofpressure-driven inflow, we add an oil well and assume a constant reservoirpressure as the boundary condition (see Fig. 2.8). Moreover, [72] suggeststhat the origin of severe-slugging instability is at the bottom-hole of thewell and that the pressure at this position is the best controlled variable.Considering the oil well dynamic is also helpful to study this possibilitytheoretically.
2.4.1 Simplified six-state model
We add two state variables, the mass of gas and mass of liquid inside theoil well, to the pipeline-riser system in (2.1a)–(2.1d) to obtain a six-statemodel. The two additional state equations are as follows.
dmg,w
dt=
(
gor
gor + 1
)
wr −wg,wh, (2.43)
dml,w
dt=
(
1
gor + 1
)
wr −wl,wh, (2.44)
where gor is the average mass ratio of gas and liquid produced from thereservoir, which is assumed to be a known parameter of the well. wg,wh andwl,wh are the flow rates of gas and liquid at the well-head. The productionrate wr from the reservoir to the well is assumed to be described by a linearrelationship.
wr = PImax(0, Pres − Pbh), (2.45)
where PI is the productivity index of the well, Pres is the reservoir pressure,which can be assumed constant in a short period of time (e.g. few months),and Pbh is the bottom-hole pressure of the well,
Pbh = Pwh + ρm,wgLw +∆Pfw. (2.46)
Here ∆Pfw is the pressure loss due to friction in the well, which is assumedto be given as
∆Pfw =αl,wλwρlU
2sl,wLw
2Dw. (2.47)
Furthermore, we have average liquid volume fraction inside well:
αl,w =ml,w
Vwρl(2.48)
38 SIMPLIFIED DYNAMICAL MODELS
Friction factor of well using same correlation as for pipeline [26]:
1√λw
= −1.8 log10
[
(
ǫ/Dw
3.7
)1.11
+6.9
Rew
]
(2.49)
Reynolds number for flow in well:
Rew =ρlUsl,wDw
µ(2.50)
Average superficial velocity of liquid in well:
Usl,w =4wnom
πDw2ρl
(2.51)
where wnom is a priori know nominal flow rate of the well. The otherimportant variables in the well model consist of the average density of thetwo-phase mixture
ρm,w =mg,w +ml,w
Vw, (2.52)
the density of the gas phase
ρg,w =mg,w
Vw −ml,w/ρl, (2.53)
and the pressure at the well-head, assuming ideal gas
Pwh =mg,wRTwh
Mg(Vw −ml,w/ρl). (2.54)
In order to calculate the volume fractions at the top of the well, we use thesame assumptions as for the phase fraction of the riser in Section 2.2.8.
αl,wt = 2αl,w − αl,wb (2.55)
In this case, because of the high pressure at the bottom-hole, the fluid fromthe reservoir is saturated [3] and liquid volume fraction at the bottom isαl,wb = 1. The gas mass fraction at top of the well is then
αmg,wt =
(1− αl,wt)ρg,wαl,wtρl + (1− αl,wt)ρg,w
. (2.56)
Density of mixture at top of the well:
ρwt = αl,wtρl + (1− αl,wt)ρg,w. (2.57)
2.4. Well-Pipeline-Riser System 39
Valve equation for subsea choke valve:
wwh = Cv3z3√
ρwtmax(Pwh − Pin, 0), (2.58)
where Pin is the pressure at the inlet of the pipeline which is given byequation (2.15) in the pipeline-riser model, Section 2.2. The flow rates ofgas and liquid phases from the well-head are as follows.
wg,wh = wwhαmg,wt (2.59)
wl,wh = wwh(1− αmg,wt) (2.60)
Flow rates of gas and liquid phases into the pipeline are respectively
wg,in = wg,wh + d1, (2.61)
wl,in = wl,wh + d2, (2.62)
where d1 and d2 can be assumed as disturbances from the other oil wells inthe network.
2.4.2 Comparison to OLGA model
In the OLGA reference model introduced in Section 2.3.1, constant inflowrates were assumed. We modifed the OLGA reference model by connectingan oil well to the inlet of the pipeline as shown in Fig. 2.8. The oil wellis vertical, has a depth of 3000 m and the same inner diameter as for thepipeline 0.12 m. The reservoir pressure is constant at 230 bar. Otherparameters related to the pipeline and the riser are same as for the OLGAreference model.
The well-pipeline-riser model includes an additional tuning parameterCv3, the valve constant of the subsea choke valve. Hence, we have five tuningparameters in the simple well-pipeline-riser model. Numerical values for allparameters are given in Table 2.1. The resulting bifurcation diagrams of thesimple model are compared to the modified OLGA model in Fig. 2.9. Thesimple model could predict the steady-state and the bifurcation point witha good accuracy. Fig. 2.9.b shows that the inlet mass flow is increasing byopening the topside choke valve. This is because of pressure-driven natureof the flow.
40 SIMPLIFIED DYNAMICAL MODELS
Pwh
Pbh
Pin
wwh
wout Z3
Z1
Figure 2.8: Schematic presentation of well-pipeline-riser system
0 20 40 60 80 100270
275
280
285
290
a) Bottom−hole pressure (Pbh
)
valve opening, Z1 [%]
Pb
h [bar
]
0 20 40 60 80 1008
10
12
14
16
b) Inlet mass flow rate (win
)
valve opening, Z1 [%]
win
[kg/
s]
0 20 40 60 80 10065
70
75
80
c) Pressure at inlet of pipeline (Pin
)
valve opening, Z1 [%]
Pin
[bar
]
0 20 40 60 80 10050
55
60
d) Pressure at top of riser (Prt)
valve opening, Z1 [%]
Prt [b
ar]
0 20 40 60 80 1000
10
20
30
40
e) Outlet mass flow rate (wout
)
valve opening, Z1 [%]
wo
ut [k
g/s]
0 20 40 60 80 1000
20
40
60
valve opening, Z1 [%]
Qo
ut [L
/s]
d) Volumetric flow rate (Qout
)
Figure 2.9: Bifurcation diagrams of simplified well-pipeline-riser model(solid lines) compared to OLGA reference model (dashedlines)
2.5. Summary 41
2.5 Summary
We have proposed a simplified dynamic model for severe-slugging flow inpipeline-riser systems. The new model and five other models from the liter-ature have been compared with a test case in the OLGA simulator. Further-more, we verified the new model experimentally. The new model cmpareswell with the OLGA simulations and the experiments.
We conclude that the proposed model maintains a good fit for steady-state and dynamics; therefore it will be used in our future works for control-lability analysis and controller design. Also, the De Meglio model is quitesimple and easy to use, and it can be considered as an alternative.
Finally, we extended the four-state model to a well-pipeline-riser systemby adding two states. The extended model was compared well to an OLGAtest case. The Matlab codes for the models are available at home page ofSigurd Skogestad.
42 SIMPLIFIED DYNAMICAL MODELS
Table 2.1: Parameters of pipeline-riser and well-pipeline-riser systems
Symb. Description Values Units
R universal gas constant 8314 J/(kmol.K)g gravity 9.81 m/s2
µ viscosity 1.426 × 10−4 Pa.sρl liquid density 832.2 kg/m3
Mg gas molecular weight 20 grPres reservoir pressure 320 barPI productivity index 2.75e-6 kg/(s.Pa)
wnom nominal mass flow from reservoir 9 kg/sgor mass gas oil ratio 0.0417 –Tw well temperature 369 KVw well volume 33.93 m3
Dw well diameter 0.12 mLw well depth 3000 mTp pipeline temperature 337 KVp pipeline volume 48.63 m3
Dp pipeline diameter 0.12 mLp pipeline length 4300 mTr riser temperature 298.3 KVr riser volume 3.14 m3
Dr riser diameter 0.1 mLr riser length 300 mLh length of horizontal section 100 mPs separator pressure 50.1 barKh level correction factor 0.7 –Kg orifice of gas flow at low-point 3.49× 10−2 –Kl orifice of liquid flow at low-point 2.81× 10−1 –Cv1 production choke constant 1.16× 10−2 –Cv3 wellhead choke valve constant 3.30× 10−3 –
2.5. Summary 43
Tab
le2.2:
Com
parison
ofdifferentsimplified
modelsto
OLGA
reference
model
Para
mete
rsOLGA
Model
Sto
rkaas
Model
Eikre
mM
odel
Kaasa
Model
Nydal
Model
DiM
eglio
Model
New
Model
Sta
teequations
Many
3diff.1alg.
4diff.
3diff.
4diff.
4diff.
4diff.
Complexity
Complicated
Average
Sim
ple
Verysimple
Average
Sim
ple
Average
Tuningpara
mete
rsMany
53
73
54
RHP
zero
sof
PrtatZ
=5%
—0.0146
0.006+
0.005i0.006–
0.005i
—0.0046
0.019+
0.034i0.019–
0.034i
0.0413
0.0126
Valuesfrom
OLGA
simulato
rErrorofth
esimplified
models
with
resp
ectto
OLGA
Criticalvalve
opening
5%
0(0%)
0(0%)
0(0%)
0(0%)
0(0%)
0(0%)
Period
[min]at
Z=
5%
15.6
0(0%)
26.15(168%)
0(0%)
15.97(102%)
0(0%)
0(0%)
Ste
pre
sponse
ofPrt
undershoot
-0.098
0.10(99%)
0.08(78%)
0.12(89%)
0.07(68%)
0.05(54%)
overshoot
0.198
0.16(80%)
0.18(89%)
—0.10(50%)
0.17(84%)
0.10(51%)
att=
10min
-0.824
0.43(53%)
0.62(75%)
0.25(30%)
0.47(58%)
0.33(41%)
Ste
ady-sta
tePin
[bar]
68.22
1.9
(2.7%)
4.4
(6.46%)
—1.77(2.6%)
0.70(1%)
0.69(1%)
Prb[bar]
66.76
——
0.02(0.04%)
——
—Prt[bar]
50.10
0.01(0.02%)
0.01(.02%)
—0.01(.02%)
0.01(0.02%)
0.01(.02%)
wout[kg/s]
9.00
0(0%)
0(0%)
—2.90(32%)
0(0%)
0(0%)
Minim
um
Pin
[bar]
63.50
2.7
(4.3%)
9(14.2%)
—8.3
(13%)
2.6
(4.1%)
2.7
(4.2%)
Prb[bar]
62.08
——
2.57(4.1%)
——
—Prt[bar]
50.09
4e-4(8e-4%)
5e-4(9e-4%)
—0.41(0.82%)
0.003(.006%)
4e-4(8e-4%)
wout[kg/s]
0.791
0.55(69%)
0.14(17%)
—0.79(100%)
3.3
(405%)
0.55(68%)
Maxim
um
Pin
[bar]
75.83
1.4
(1.8%)
1.89(2.5%)
—1.00(1.3%)
1.3
(1.7%)
1.50(2.0%)
Prb[bar]
74.55
——
0.55(.075%)
——
—Prt[bar]
50.14
1.5
(3%)
0.95(1.9%)
—1.09(2.2%)
0.71(0.35%)
0.33(0.16%)
wout[kg/s]
31.18
80(278%)
57(200%)
—13.3
(43%)
22(77%)
9.10(32.0%)
44 SIMPLIFIED DYNAMICAL MODELS
Table 2.3: Parameters of small-scale experimental setup
Symb. Description Values Units
µ viscosity 1.426 × 10−4 Pa.sρl liquid density 832.2 kg/m3
Mg gas molecular weight 18 grTp pipeline temperature 288 KVp pipeline volume 0.0219 m3
Dp pipeline diameter 0.02 mLp pipeline length 69.71 mTr riser temperature 288 KVr riser volume 0.001 m3
Dr riser diameter 0.02 mLr riser length 3 mLh length of horizontal section 0.2 mPs separator pressure 0.013 barKh level correction factor 1 –Kg orifice of gas flow at low-point 2.07 × 10−2 –Kl orifice of liquid flow at low-point 1.57 × 10−1 –Cv1 production choke constant 2.21 × 10−4 –
Chapter 3
CONTROLLED VARIABLE
SELECTION
The focus of this chapter is to design simple, yet robust structures for anti-slug control systems (stabilizing control). Control structure design is toidentify suitable controlled variables and manipulated variables and theirpairings for control. To this end, we perform a controllability analysis of thesystem with different available measurements and alternative manipulatedvariables.
The controllability is often evaluated by considering the minimum achiev-able peaks of individual closed-loop transfer functions (e.g the sensitivity Sof complementary sensitivity T ). These bounds are independent of thecontroller design and they are physical properties of the plant (process).Controlled variables, or a combinations of these, that result in small peaksare preferable. We extend the controllability analysis to a mixed sensitivityH∞ optimization problem, and we introduce the γ-value as a single measureof controllability. The aim is to unify different and sometimes conflictingcontrollability measures.
However, our controllability analysis is based on linear systems. Never-theles, although the nature of severe-slugging, and also the simplified modelused in this work, is highly nonlinear, we found that the controllabilityanalysis gives useful information about the fundamental limitations.
Two case studies have been considered. First, a controllability analysisis performed for a well-pipeline-riser system using a 6-state model. Then,the casing-heading instability of gas-lifted oil wells is considered. The resultsprovided in this chapter have been presented in [39], [40].
45
46 CONTROLLED VARIABLE SELECTION
3.1 Controllability: Theoretical Background
ym
Ku y
G
Gd
++
d
+
+
+
n
–r
Figure 3.1: Feedback control loop
The conventional state controllability of Kalman [50] is not our interestin this work; instead the more practical concept of input-output controlla-bility as defined in [74] is used.Definition 1 : (Input-output) controllability is the ability to achieve ac-ceptable control performance; that is, to keep outputs (y) within specifiedbounds or displacement from their references (r), in spite of unknown butbounded variations, such as disturbances (d) and plant changes (includinguncertainty), using available inputs (u) and available measurements (ym anddm).
The controllability analysis is performed at unstable operating points,with the valve opening Z1 at 10% and 20%.
3.1.1 Closed-loop transfer functions
We consider a linear process model in the form y = G(s)u +Gd(s)d with afeedback controller u = K(s)(r−y−n) where d represents disturbances andn is the measurement noise (Figure 3.1). The resulting closed-loop responseis
y = Tr + SGdd− Tn, (3.1)
or
e = r − y = Sr − SGdd+ Tn, (3.2)
where S = (I + GK)−1 and T = GK(I + GK)−1 = I − S represent thesensitivity and the complementary sensitivity functions, respectively.
For an input disturbance, we have Gd = G, so the transfer function SGis related to the effect of input disturbances on the control error r− y. The
3.1. Controllability: Theoretical Background 47
control input for the closed-loop system is
u = KS(r −Gdd− n). (3.3)
Therefore, we see that the transfer function KS is very important whenevaluating input usage (u). One should notice that the closed-loop transferfunctions S, T,KS and SG can also be regarded as the measures of robust-ness against different types of uncertainty [77]. These transfer functionsshould be as small as possible to achieve better robustness properties ofthe control system. For instance, the sensitivity transfer function S is alsothe sensitivity to inverse relative uncertainty, which is a good representa-tion of uncertainty in the pole locations [74]. In summary, the magnetudeof the closed-loop transfer functions S, T,KS, SG,KSGd and SGd provideinformation regarding both achievable performance and robustness.
By “peak” we mean the maximum value of the frequency response orH∞ norm, ‖M‖∞ = max
ω‖M(jω)‖. The minimum acheiavable peaks are
denoted MM,min = minK
‖M‖∞ (e.g. MS,min = minK
‖S‖∞ and MT,min =
minK
‖T‖∞). The bounds presented in the following are independent of the
controller K, and they are the physical properties of the process itself. Thebounds are, however, dependent on a systematic and correct scaling of thevariables as explained later.
3.1.2 Lower bound on S and T
The lowest achievable peak of the sensitivity function, MS,min, is determinedby the distance between the unstable (RHP) poles (pi) and zeros (zi) of theprocess. For SISO systems, we have the following bound for any unstable(RHP) zero z [74]:
MS,min =
Np∏
i=1
|z + pi||z − pi|
. (3.4)
We note that the bound increases rapidly as the unstable zero z gets closeto an unstable pole pi. The bound is tight for a plant with a single RHPzero. The lowest achievable peak of the complementary sensitivity functionis similarly bounded
MT,min =
Nz∏
j=1
|zj + p||zj − p| .|e
pθ| (3.5)
and we also have a penalizing term epθ for the time delay θ. In (3.5) zjdenotes the Nz RHP-zeros of G(s) and θ denotes the time delay of G(s).
48 CONTROLLED VARIABLE SELECTION
The bound in (3.5) is tight for plants with a single RHP pole. Ofcourse,the achievable peak of S also increases for an unstable plant with timedelay, because of the identity constraint S + T = I, which implies that|MS,min| ≥ |MT,min| + 1, but we do not have a tight bound for MS,nmin
when threre is a time delay. For MIMO systems with no time delays, thefollowing general tight bounds apply for any number of RHP-poles andRHP-zeros [8]:
MS,min = MT,min =
√
1 + σ2(Q−1/2p QzpQ
−1/2z ), (3.6)
Here, the elements of the matrices Qz, Qp and Qzp are given by [8]
[Qz]ij =yHz,iyz,j
zi + zj, [Qp]ij =
yHp,iyp,j
pi + pj, [Qzp]ij =
yHz,iyp,j
zi − pj(3.7)
The vectors yz,i and yp,i are the (unit) output direction vectors of the zerozi and pole pi, respectively.
3.1.3 Lower bound on KS
The transfer function KS gives the effect of the measurement noise n andoutput disturbances on the plant input u. For SISO systems, its lowestachievable peak can be calculated from the ([27], [28])
MKS,min = |Gs(p)−1|, (3.8)
where Gs is the stable version of G with the RHP-poles of G mirrored intothe LHP. The bound is tight (with equality) for an plant with one realunstable pole p. For MIMO plants with any number of unstable poles pi, atight bound is [22]
MKS,min = 1/σH(U(G)∗), (3.9)
where σH is the smallest Hankel singular value and U(G)∗ is the mirrorimage of the antistable part of G. For a stable plant there is no lowerbound.
3.1.4 Lower bounds on SG and SGd
The transfer function SG should to be small to reduce the effect of inputdisturbances on the control error, and also for robustness against pole un-certainty. SGd is related to the effect of a general disturbance. The twofollowing bounds apply for an unstable zero z in G [74]:
MSG,min = |Gms(z)|Np∏
i=1
|z + pi||z − pi|
, (3.10)
3.1. Controllability: Theoretical Background 49
MSGd,min = |Gd,ms(z)|Np∏
i=1
|z + pi||z − pi|
, (3.11)
Here, Gms and Gd,ms are the “minimum-phase, stable version” of the trans-fer functions G and Gd, respectively, with both RHP-poles and RHP-zerosmirrored into LHP. These bounds are tight for a single unstable zero z, butsince they are valid for any RHP-zero z, they are also usefull for systemswith multiple unstable zeros [77].
3.1.5 Lower bound on KSGd
The bound in (3.9) can be generalized [74]
MKSGd,min = 1/σH(U(G−1d,msG)∗). (3.12)
where U(G−1d,msG)∗ is the mirror image of the anti-stable part of G−1
d,msG.This bound is tight for multiple and complex unstable poles pi. Note thatany unstable modes in Gd must be contained in G such that they are stabi-lizable with feedback control. A simpler bound is obtianed by using equation(3.8) for any unstable pole p:
MKSGd,min = |G−1s (p)|.|Gd,ms(p)|. (3.13)
This bound is tight only for SISO systems with one real unstable pole p[27], [74].
3.1.6 Pole vectors
The output pole vector yp,i for a process with state-space representation(A,B,C,D) is defined as [29]
yp,i = Cti, (3.14)
where ti is the right (normalized) eigenvector associated with pi (Ati = piti).Based on minimum input usage for stabilization, it can be suggested thatthe measurements with the largest element in the output pole vector shouldbe used for stabilizing control [29]. In the same way, for input selection,the input that has the largest element in the input pole vector up,i = BHqi,where qi that is the left eigenvector of A (qHi A = piq
Hi ) should be used.
One limitation on the use of pole vectors is that the relationship betweenthe magnitude of the input usage and the magnitude of the pole vectorselements only hold for a plant with a single unstable pole p. In our system,there is a pair of complex conjugate unstable poles pi, but pole vectors stillgive useful information about the measurement selection [77].
50 CONTROLLED VARIABLE SELECTION
3.1.7 Mixed sensitivity controllability analysis
The above controllability measures were also considered in the previousworks [71], [77]. One limitation is that these measures consider only onetransfer function at a time, and may give conflicting results. To get a singlemeasure (γ), we consider an H∞ problem where we want to bound σ(S) forperformance, σ(T ) for robustness and low sensitivity to noise, and σ(KS) topenalize large inputs. These requirements may be combined into a stackedH∞ problem [74].
minK
‖N(K)‖∞, N
∆=
WuKSWTTWPS
(3.15)
where Wu, WT and WP determine the desired shapes of KS, T and S,respectively. Typically, W−1
P is chosen to be small at low frequencies toachieve good disturbance attenuation (i.e., performance), andW−1
T is chosento be small outside the control bandwidth, which helps to ensure goodstability margin (i.e., robustness). Wu is often chosen as a constant. Thesolution to this optimization problem gives a stabilizing controller K thatsatisfies [16], [23]:
σ(KS(jω)) ≤ γσ(W−1u (jω))
σ(T (jω)) ≤ γσ(W−1T (jω))
σ(S(jω)) ≤ γσ(W−1P (jω))
(3.16)
We want to compare choices for controlled variables (CVs). To have thesame cost function in all cases, all the candidate CVs are included in the y1part and the particular CV for evaluation is in the y2 part of the generalizedplant in Figure 3.2. The value of γ in equation (3.16) should be as small aspossible for good controllability.
3.1.8 Low-frequency performance
Disturbance rejection is not the main objective for stabilizing control, but toavoid the possible destabilizing effect of nonlinearity, the system should not“drift” far away from its nominal operating point. For disturbance rejectionwithout input saturation, we need |G(jω)| ≥ |Gd(jω)| at the frequencieswhere |Gd| > 1. In particular, to achieve low-frequency performance, thesteady-state gain of the plant must be large enough [77].
3.1.9 Scaling
One important step before performing a controllability analysis is to scaleinputs, outputs and disturbances of the plant. In Definition 1, the bound
3.2. Well-Pipeline-Riser System 51
y1
y2
K(s)
G(s)
WP
Wu
WTy
GENERALIZED PLANT P(s)
eu1
u2
+_
CONTROLLER
u
Figure 3.2: Closed-loop system for mixed sensitivity control design
that the controlled variable must be kept within is not the same for differentcontrolled variables. For a correct comparison between candidate controlledvariables, each CV must be scaled based on its maximum allowed variations,that the maximum allowed is in the interval [-1,1] for all CVs. We do this bydividing each CV by a scaling factor Dy as given in Table 3.1 and Table 3.3.Disturbances are also scaled to be in the range of [-1,1]. The maximumexpected deviations of inflow rates (disturbances) are 10% of the nominalvalues of wg,in and wl,in. This gives the following scaling weights in firstcase study:
Dd =
[
1 00 0.04
]
The inputs are scaled similarly. The controllability analysis is performedat two operating points (Z1 = 10% and Z1 = 20%); and the maximumpossible change of u at the two operating points are therefore Du = 0.1 andDu = 0.2 respectively.
3.2 Well-Pipeline-Riser System
In this Section, we consider the topside valve opening Z1 as the manipulatedvariable and we find suitable candidate controlled variables.
3.2.1 Summary of simplified model
First, we considered the pipeline-riser system with constant inflow rates. APDE-based two fluid model with 13 segments was used for the controllability
52 CONTROLLED VARIABLE SELECTION
analysis in [77]. This model resulted in a set of 50 ODEs. In their work,it was concluded that main dynamics of severe slugging in a pipeline-risersystem can be captured by a simpler model and we proposed the four-statesimplified model given in Section 2.4. Then, we extended this model to awell-pipeline-riser system by adding an oil well as the boundary conditions(Figure 3.3). The oil well dynamics are modelled by two additional statevariables, representing the masses of gas and liquid in the well. The stateequations of the six-state model are as follows:
dmg,w
dt=
(
gor
gor + 1
)
wr − wg,wh (3.17a)
dml,w
dt=
(
1
gor + 1
)
wr − wl,wh (3.17b)
dmg,p
dt= wg,in − wg,rb (3.17c)
dml,p
dt= wl,in − wl,rb (3.17d)
dmg,r
dt= wg,rb −wg,out (3.17e)
dml,r
dt= wl,rb − wl,out (3.17f)
See Section 2.4 for details of this augmented model. The advantage withthis case is that inflow rates are pressure-driven which is closer to practicalconditions. In addition, we can consider measurements in the oil well foranti-slug control as suggested in [72]. The OLGA case for the well-pipeline-riser systems introduced in Section 2.4.2 is used a reference for the modelfitting. We fit the simplified model to the OLGA case by adjusting fiveparameters in the model. The parameter values found by trial and error aregiven in Table 2.1.
3.2.2 Bounds on minimum achievable peaks
Different candidate controlled variables of the well-pipeline-riser system areshown in Figure 3.3. The minimum achievable peaks for selected closed-loop transfer functions are given in Table 3.1 and 3.2 for two operatingpoints Z1 = 10% and Z1 = 20%, receptively. The minimum peaks of Tfor Prt, ρrt and αl,t in Table 3.1 are relatively large, and it is expected tobe difficult to use these measurements as controlled variables. The largepeaks are because of RHP-zeros in transfer functions for these variables.
3.2. Well-Pipeline-Riser System 53
Pwh
Pbh
Pin
Prt , rt , l,rt
win
wout
Qout
Prb
Z3
Z1Candidate MVs
Z1: opening of topside valve
Z2: opening of riser-base valve
Z3: opening of well-head valve
Candidate CVs
Pbh: pressure at bottom-hole of well
Pwh: pressure at well-head
win: inlet mass flow rate to pipeline
Pin: pressure at pipeline inlet
Prb: pressure at riser base
DPr: pressure drop over riser (Prt - Prb)
Prt: pressure at top of riser
Qout: top-side choke volumetric flow rate
wout: top-side choke mass flow rate
rt : mixture density at top
l,rt: liquid volume fraction at top
Z2
Figure 3.3: Schematic presentation of candidate controlled variables andmanipulated variable of well-pipeline-riser system
The RHP-zero dynamics are observed as inverse response in the step testof the system.
The location of RHP-poles of the process and the RHP-zeros for Prt areplotted as a function of the valve opening in Figure 3.4. For Z1 = 5% theprocess is on the limit to intability whith a pair of complex conjugate poleson the Jω-axis. For Z1 > 5% the two complex poles are in the RHP, and theprocess is unstable. For larger valve openings, the unstable RHP-poles andthe RHP-zero get very close to each other. From equations (3.4) and (3.5),we see that when the RHP-pole and the RHP-zero are close to each other,the peak of the sensitivity and complementary sensitivity transfer functionsare large. This is also seen from Table 3.1 and Table 3.2; the minimum peakfor T of Prt is much larger for Z1 = 20% than for for Z1 = 10%.
One of RHP-poles of the system moves far away from the Jω-axis asthe valve opening increases. The faster unstable dynamics also make thestabilization more difficult. Another increasing limitation when openingthe valve is that gain of the system decreases. These effects can be seenby comparing peack of KS and G(0) for Z1 = 10% and Z1 = 20%. Wheresmaller gain shows that we need to increase the controller gain to stabilizethe process and the higher peak shows that a more aggressive control actionis needed for larger valve openings for all CV candidates. It is desired tocontrol the system with a larger valve opening to achieve a higher production
54 CONTROLLED VARIABLE SELECTION
rate, but it is not possible because of fundamental controllability limitations.These findings are in agreement with the results reported by [77] where amore detailed two-fluid model is used. Unlike the pipeline-riser system([77]), Qout and wout show considerable steady-state gains for the presentwell-pipeline-riser system which is a result of the pressure-driven nature onthe inflows. Therefore, flow rates also can be used in a single loop for thestabilizing control without the drift problem.
The minimum achievable peak of T for DPr is 1, and no problem interm of controllability is expected. However, the small value of the steady-state gain G(0) implies that the system might drift away from the desiredoperating point. The controllability data for the combined measurementsin Table 3.1 and Table 3.2 show that combining one pressure measurementand one flow rate gives the best result in terms of controllability.
In parctice, there may be time delays in the system, for example, relatedto the measurements, or valves. From (3.5) the bound on the complementarysensitivity function increase by a factor epθ. In OLGA simulations, it wasobserved that the time delay for the measurement from the inlet of thepipeline and the oil well is θ = 15 sec. The factor epθ becomes 1.19 forZ1 = 10% and 1.29 Z1 = 20%, because the unstable pole p is larger inmagnitude for large valve openings (Figure 3.4). The bottom-hole pressuredemonstrates the largest steady-state gain and relatively small values forminimum achievable peaks of all closed-loop transfer functions. This meansthat the pressure at the bottom-hole is an effective controlled variable.
3.2.3 Control structure selection
Next, we consider the mixed sensitivity analysis, which aims to combineimportant controllability measures into one (γ), where a small value of γis desired. The resulted γ values given in Table 3.1 and 3.2 contain theinformation of the other peaks given for each CV. We emphesize role of theγ values by using them as a measure for CV slection. In addition, we showsimulation results of the corresonding H∞ controller for selected cases. Allsimulations presented in this paper are based on scaled variables, and theideal is to keep the controlled variables in the range of [-1,1]. For the well-pipeline-riser case study, we added 10 sec time delay at input of the systemin all simulations to test robustness of the different control structures.
For Z1 = 10%, the best single measurement apprears to be Pbh (γ =20.70). Figure 3.5 shows the simulations using the obtained H∞ controllerfor this case. For the top-side pressure Prt we achieved γ = 35.48 whichshows a poor robustness in Figure 3.5, but the outlet flow Qout (γ = 28.35)
3.2. Well-Pipeline-Riser System 55
Tab
le3.1:
Con
trollabilitydataforwell-pipeline-risercase
studyat
operatingpointZ1=
10%
CV
Value
Dy
G(0)
Pole
vector
MT
MK
SM
SG
MK
SG
d1
MK
SG
d2
MSG
d1
MSG
d2
γ
Pbh[bar]
281.95
1-3.80
0.0065
1.19
0.62
00.24
0.46
00
20.70
Pwh[bar]
68.85
1-2.76
0.0048
1.19
0.86
00.24
0.46
00
28.23
win[kg/s]
10.46
11.05
0.0037
1.19
1.10
00.24
0.46
00
31.02
Pin[bar]
68.66
1-2.80
0.0049
1.19
0.84
00.24
0.46
00
27.33
Prb[bar]
67.26
1-3.07
0.0059
10.70
00.23
0.56
00
21.21
DPr[bar]
15.54
1-0.18
0.0088
1.02
0.47
0.03
0.19
0.46
0.033
0.020
89.07
Prt[bar]
51.72
1-2.89
0.0039
3.17
1.05
5.30
0.17
0.54
0.25
0.66
35.48
Qout[L/s]
20.66
21.24
0.0115
10.36
00.24
0.53
00
28.35
wout[kg/s]
10.46
11.05
0.0140
10.29
00.16
0.57
00
31.02
ρrt[kg/m
3]
506.56
50
-0.23
0.0055
3.43
0.74
5.50
0.30
0.84
0.69
1.70
85.47
αl,rt[−
]0.58
1-0.013
0.0004
3.43
11.70
0.35
0.30
0.84
0.044
0.10
–Pin&Prt
––
––
10.65
00.22
0.57
00
16.72
Pbh&Q
out
––
––
10.29
00.07
0.04
00
8.93
Pin&Q
out
––
––
10.33
00.22
0.76
00
12.07
Prt&Q
out
––
––
10.34
00.06
0.04
00
12.64
Prt&ρrt
––
––
10.61
00.23
0.72
00
22.84
56CONTROLLEDVARIABLESELECTION
Tab
le3.2:
Con
trollabilitydataforwell-pipeline-risercase
studyat
operatingpointZ1=
20%
CV
Value
Dy
G(0)
Pole
vector
MT
MK
SM
SG
MK
SG
d1
MK
SG
d2
MSG
d1
MSG
d2
γ
Pbh[bar]
280.25
1-1.15
0.0045
1.28
3.06
00.58
0.95
00
–Pwh[bar]
67.69
1-0.83
0.0033
1.28
4.24
00.58
0.95
00
–w
in[kg/s]
10.93
10.32
0.0044
1.28
2.69
00.58
0.95
00
98.3
Pin[bar]
67.48
1-0.84
0.0035
1.28
3.89
00.58
0.95
00
–Prb[bar]
65.98
1-0.92
0.0060
12.12
00.58
0.95
00
152.3
DPr[bar]
15.43
1-0.056
0.0087
1.03
1.35
0.02
0.58
0.95
0.034
0.016
104.0
Prt[bar]
50.55
1-0.87
0.0010
12.30
13.63
7.05
0.56
1.44
0.32
0.82
730.5
Qout[L/s]
21.79
20.37
0.0122
11.07
00.58
1.20
00
67.30
wout[kg/s]
10.93
10.32
0.0125
11.15
00.56
1.73
00
72.55
ρrt[kg/m
3]
501.63
50
-0.094
0.0012
14.74
15.23
8.74
5.22
10.69
2.76
6.10
–αl,rt[−
]0.58
1-0.006
0.0000
14.64
238.0
0.55
5.19
10.64
0.17
0.39
–Pin&Prt
––
––
12.53
00.93
2.04
00
110.7
Pbh&Q
out
––
––
10.89
00.93
1.88
00
28.46
Pin&Q
out
––
––
10.90
01.57
3.46
00
36.77
Prt&Q
out
––
––
10.86
00.93
1.88
00
39.36
Prt&ρrt
––
––
10.77
02.78
5.00
00
–
3.2. Well-Pipeline-Riser System 57
−0.01 0 0.01 0.02 0.03 0.04 0.05−0.015
−0.01
−0.005
0
0.005
0.01
0.015
Real axis
Imag
inar
y ax
is
Z=1%Z=90%
Z=1% Z=90%
Z=5%
Z=10%
Z=20%
Z=30%
Z=40% Z=90%
Z=5%
Z=10%
Z=20%
Z=30%
Z=40%
Z=90%
RHP−ZerosRHP−poles
Figure 3.4: Location of RHP-poles and RHP-zeros of top pressure of well-pipeline-riser system for different valve openings
is quite promising in the simulation results in Figure 3.7. We see that theoutlet flow can be used for the anti-slug control in theory.
However, it is better to combine measurements, and the combination ofone upstream pressure and the outlet flow results in the smallest value for γ.Simulation result with a H∞ control (γ = 8.93) using Pbh and Qout as thecontrolled variables is illustrated in Figure 3.8. Even if the subsea pressuremeasurements are not available, combing the top-side pressure Prt with theoutlet flow Qout gives a statisfactory result (γ = 12.64). This is shown bythe simulation results in Figure 3.9.
Remarks
It should be noted that for combing one flow rate and one pressure, it isimpossible to have tight control of both at the same time. If tight pressurecontrol is required, flow rate can not be controlled tightly. Because, when aninflow disturbance comes to the system, it needs to be released to maintaina constant pressure. In this situation, the flow control can only help forrobustness of the pressure control. Figure 3.8 and Figure 3.9 show thiscondition where outlet flow Qout is not controlled tightly. This performancerequirement was simulated by considering integral action (i.e. small valuefor W−1
P at low frequency) for control of the pressure (Pbh in Figure 3.8and Prt in Figure 3.9) and no integral action for Qout (i.e. the weight
58 CONTROLLED VARIABLE SELECTION
0 2 4 6 8−1
0
1
Z1
0 2 4 6 8−1
0
1
d1
d2
0 2 4 6 8−2
0
2
Qout
0 2 4 6 8−2
0
2
setpoint Pbh
0 2 4 6 8−2
0
2
time [h]
Pin
0 2 4 6 8−2
0
2
time [h]
Prt
Figure 3.5: H∞ control of well-pipeline-riser, CV = Pbh, MV = Z1
of W−1P = 1/Ms in which Ms is the desired peak of sensitivity transfer
function). Cascade control with the flow control as the inner loop is asimple structure to implement this case in practice. On the other hand, iftight control on the flow rate is required, the pressure fluctuations becauseof disturbances are unavoidable.
3.3 Gas-Lifted Oil Well
3.3.1 Summary of simplified model
We use a simplified three-state model of the system for the controllabilityanalysis. The model is very similar to the one used in [1], [2], [18], [19], [20].The state variables are the mass of gas in the annulus (mg,a), the mass ofgas in the tubing (mg,t) and the mass of liquid in the tubing (ml,t). Thestate equations are as follows
dmg,a
dt= wg,in − wg,inj (3.18a)
dmg,t
dt= wg,inj + wg,res − wg,out (3.18b)
dml,t
dt= wl,res −wl,out (3.18c)
3.3. Gas-Lifted Oil Well 59
0 2 4 6 8−1
0
1
Z1
0 2 4 6 8−1
0
1
d1
d2
0 2 4 6 8−2
0
2
Qout
0 2 4 6 8−2
0
2
Pbh
0 2 4 6 8−2
0
2
time [h]
Pin
0 2 4 6 8−2
0
2
time [h]
setpoint Prt
Figure 3.6: H∞ control of well-pipeline-riser, CV = Prt, MV = Z1
0 2 4 6 8−1
0
1
Z1
0 2 4 6 8−1
0
1
d1
d2
0 2 4 6 8−2
0
2
setpoint Qout
0 2 4 6 8−2
0
2
Pbh
0 2 4 6 8−2
0
2
time [h]
Pin
0 2 4 6 8−2
0
2
time [h]
Prt
Figure 3.7: H∞ control of well-pipeline-riser, CV = Qout, MV = Z1
60 CONTROLLED VARIABLE SELECTION
0 2 4 6 8−1
0
1
Z1
0 2 4 6 8−1
0
1
d1
d2
0 2 4 6 8−2
0
2
setpoint Qout
0 2 4 6 8−2
0
2
setpoint Pbh
0 2 4 6 8−2
0
2
time [h]
Pin
0 2 4 6 8−2
0
2
time [h]
Prt
Figure 3.8: H∞ control of well-pipeline-riser, CV = [Pbh, Qout], MV =Z1
0 2 4 6 8−1
0
1
Z1
0 2 4 6 8−1
0
1
d1
d2
0 2 4 6 8−2
0
2
setpoint Qout
0 2 4 6 8−2
0
2
Pbh
0 2 4 6 8−2
0
2
time [h]
Pin
0 2 4 6 8−2
0
2
time [h]
setpoint Prt
Figure 3.9: H∞ control of well-pipeline-riser, CV = [Prt, Qout], MV = Z1
3.3. Gas-Lifted Oil Well 61
Gas lift
choke
Production
choke
Oil outlet
Gas inlet
Annulus
Injection
valve
Tubing
Candidate CVs
wg,in: inlet gas mass flow rate
Pat: annulus top pressure
Pab: annulus bottom pressure
Pbh: bottom-hole pressure
Ptt: tubing top pressure
wout: outlet mass flow rate
tt: mixture density at top of tubing
l,tt: liquid volume fraction at top
Candidate MVs
u1: opening of production choke
u2: opening of gas lift choke
DVs
Pres: reservoir pressure
Pgs: gas lift source pressure
Pres
Pgs
Ptt tt l,tt
Pat
Pbh
u1
u2wg,in
wout
Pab
Figure 3.10: Presentation of candidate CVs and MVs of gas-lift oil well
where wg,in is the mass flow rate of inlet gas to the annulus and wg,inj is themass flow of injected gas from the annulus into the tubing. wg,res and wl,res
are gas and liquid mass flow rates from the reservoir to the tubing. wg,out
and wl,out are the mass flow rates of gas and oil outlet from the tubing,respectively. These flow rates are calculated by additional equations givenin Appendix A. The simplified model was fitted to a test case implementedin the OLGA simulator. Constants and parameters used in the model aregiven in Table A.1.
3.3.2 Bounds on minimum achievable peaks
Figure 3.10 shows different candidate controlled variables and the two alter-native manipulated variables of the gas-lifted oil well. We identify suitableCVs for this case by using a similar controllability analysis as for the well-pipeline-riser system. The minimum achievable peaks for different closed-loop transfer functions for the gas-lifted oil well case are given in Tables 3.3,3.4 and 3.5. Location of RHP-poles of the system and RHP-zeros of Ptt foru2 = 0.4 and as a function of the production choke valve opening u1 are
62 CONTROLLED VARIABLE SELECTION
−2 0 2 4 6 8 10 12
x 10−3
−3
−2
−1
0
1
2
3x 10
−3
real axis
imag
inar
y ax
is
u1=0.1 u
1=1u
1=0.1u
1=1
u1=0.1
u1=0.3
u1=0.5
u1=0.7
u1=0.9
u1=1
u1=0.1
u1=0.3
u1=0.5
u1=0.7
u1=0.9
u1=1
RHP−zerosRHP−poles
Figure 3.11: Location of RHP-poles of system and RHP-zeros of tubingtop pressure for u2 = 0.4 and different values of u1
shown in Figure 3.11. The critical valve opening for transition from stableto unstable operation is u1 = 0.3. Two poles of the system move to the RHPand the system becomes unstable for u1 > 0.3; this is in agreement withthe stability map in Figure 3.12. The controllability results when using theproduction choke valve u1 as the manipulated variable are very similar towell-pipeline-riser in the previous Section. Ptt, ρmix,t and αL,t are not suit-able controlled variables in a single loop, because of RHP-zero dynamics.Ptt shows two RHP-zeros for all u1 values (Figure 3.11). One of the RHP-zeros does not move so much and it is always close to pole locations. As theproduction valve opening u1 increases, RHP-poles get closer to the smaller(important) RHP-zero. The large peak of the complementary sensitivitydoes not occur when Ptt combines with other measurements in Table 3.3,because the system becomes non-square and zeros disappear.
The bottom-hole pressure Pbh shows the best controllability properties.It has the largest element in the output pole vector that makes it suitablefor stabilization of the unstable system. Pbh also has the largest steady-state gain G(0) and the smallest values for all of the closed-loop transferfunctions. In the second place, the pressure at the bottom of the annulusPab shows good controllability properties. The third good candidate is thepressure at top of the annulus, Pat.
We performed a similar controllability analysis for using u2 as the single
3.3. Gas-Lifted Oil Well 63
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
production choke opening (u1)
inje
cted
gas
rat
e (w
G,in
)
Operating Point
Stable
Unstable
model stability transitionOLGA stability transition
Figure 3.12: Stability transition of system, blue markers for OLGA andred markers for simplified model
manipulating variable of the gas-lifted oil well (Table 3.4), but the resultsusing u2 were not satisfactory.
3.3.3 Control structure selection
Based on the γ values provided in Table 3.3, 3.4 and 3.5, we can makerecommendation about the control structure. First, we consider using theproduction choke valve u1 as the manipulated variables. For a SISO controlstructure, the bottom-hole pressure Pbh is the best CV from our results(γ = 3.6). This is in accordance with previous works [20], [1]. Pbh usually isnot directly measurable, but as suggested by [20] and [1], it can be estimatedusing an observer.
In simulations related to gas-lifted oil well case study, we added 20 sectime delay at input of the system in all simulations to test robustness ofthe different control structures. Simulation result of using Pbh as the singleCV is shown in Fig 3.13. Simulation results of using Ptt(γ = 19.16) andPat(γ = 14.85), shown in Fig 3.14 and Fig 3.15, respectively, indicate poorperformance when they are used for SISO control.
Using two controlled variables and one manipulated variable, it is im-possible to get tight control on the both controlled variables at the sametime. Similar to a cascade controller, we can have tight control with a con-stant set-point only on one of controlled variables. We calculated γ1 when
64CONTROLLEDVARIABLESELECTION
Tab
le3.3:
Con
trollabilitydataforgas-lift
oilwellusingu1as
man
ipulatedvariab
le
CV
Value
Dy
G(0)
Pole
vector
MT
MK
SM
SG
MK
SG
d1
MK
SG
d2
MSG
d1
MSG
d2
γ1
γ2
wgin[kg/s]
0.86
0.05
0.76
0.0004
1.00
3.04
0.00
0.23
1.987
0.00
0.00
89.55
–Pat[bar]
81.16
15.22
0.0031
1.00
0.44
0.00
0.23
0.10
0.00
0.00
14.85
–Ptt[bar]
20.89
15.72
0.0028
3.06
0.38
10.49
0.25
0.11
0.69
0.42
19.16
–Pab[bar]
90.35
15.81
0.0034
1.00
0.40
0.00
0.23
0.10
0.00
0.00
13.54
–Pbh[bar]
88.56
16.95
0.0089
1.00
0.11
0.00
0.23
0.09
0.00
0.00
3.60
–w
out[kg/s]
18.51
20.88
0.0024
1.00
0.49
0.00
0.30
0.11
0.00
0.00
19.35
–ρtt[kg/m
3]
186.96
20
1.61
0.0013
3.11
1.24
3.77
0.56
0.29
0.71
0.38
38.08
–αl,tt[−
]0.23
0.23
0.17
0.0001
3.11
10.83
0.43
0.57
0.30
0.08
0.04
289.88
–Pabw
out
––
–0.0034
1.00
0.26
0.00
0.13
0.05
0.00
0.00
8.89
19.35
Pabρtt
––
–0.0034
1.00
0.36
0.00
0.20
0.08
0.00
0.00
12.20
13.17
Pabw
g,in
––
–0.0034
1.00
0.40
0.00
0.12
0.10
0.00
0.00
13.42
22.13
PatPbh
––
–0.0089
1.00
0.11
0.00
0.12
0.05
0.00
0.00
4.52
3.45
PatPtt
––
–0.0031
1.00
0.26
0.00
0.12
0.05
0.00
0.00
8.65
11.58
Patw
out
––
–0.0031
1.00
0.27
0.00
0.13
0.05
0.00
0.00
9.17
19.35
Patρtt
––
–0.0031
1.00
0.39
0.00
0.20
0.08
0.00
0.00
13.12
13.96
Patw
g,in
––
–0.0031
1.00
0.44
0.00
0.12
0.10
0.00
0.00
14.70
22.13
Pbhw
out
––
–0.0089
1.00
0.10
0.00
0.13
0.05
0.00
0.00
3.39
19.35
Pbhρtt
––
–0.0089
1.00
0.11
0.00
0.20
0.09
0.00
0.00
3.53
10.96
Pbhw
g,in
––
–0.0089
1.00
0.11
0.00
0.12
0.10
0.00
0.00
3.60
22.13
PttPbh
––
–0.0089
1.00
0.10
0.00
0.12
0.05
0.00
0.00
7.25
3.39
Pttw
out
––
–0.0028
1.00
0.30
0.00
0.14
0.06
0.00
0.00
15.41
19.35
Pttρtt
––
–0.0028
1.00
0.34
0.00
0.21
0.10
0.00
0.00
16.97
12.23
Pttw
g,in
––
–0.0028
1.00
0.37
0.00
0.12
0.12
0.00
0.00
18.64
22.13
woutw
g,in
––
–0.0024
1.00
0.47
0.00
0.13
0.12
0.00
0.00
19.35
22.13
3.3. Gas-Lifted Oil Well 65
Tab
le3.4:
Con
trollabilitydataforgas-lift
oilwellusingu2as
man
ipulatedvariab
le
CV
Value
Dy
G(0)
Pole
vector
MT
MK
SM
SG
MK
SG
d1
MK
SG
d2
MSG
d1
MSG
d2
γ1
γ2
wgin[kg/s]
0.86
0.1
8.85
0.0002
29.30
3.21
186.46
0.22
1.45
0.96
6.75
153.71
—Pat[bar]
81.16
13.32
0.0031
1.75
0.23
6.43
0.22
0.08
1.31
0.23
21.80
—Ptt[bar]
20.89
15.60
0.0028
3.04
0.38
12.03
0.22
0.10
0.74
0.44
29.64
—Pab[bar]
90.35
13.70
0.0034
1.75
0.21
7.16
0.22
0.08
1.46
0.26
20.60
—Pbh[bar]
88.56
119.36
0.0089
2.77
0.11
36.98
0.22
0.10
2.88
1.34
14.68
—w
out[kg/s]
18.51
22.83
0.0024
3.01
0.43
10.06
0.25
0.10
0.79
0.36
32.17
—ρtt[kg/m
3]
186.96
20
3.17
0.0013
4.77
0.65
12.09
0.35
0.18
0.83
0.44
48.22
—αl,tt[−
]0.23
00.38
0.0001
4.76
5.58
1.45
0.36
0.19
0.10
0.05
402.73
—Pabρtt
––
–0.0034
1.00
0.18
0.00
0.16
0.08
0.00
0.00
18.83
16.28
Pabw
g,in
––
–0.0034
1.00
0.21
0.00
0.11
0.08
0.00
0.00
20.47
20.35
Pabw
out
––
–0.0034
1.00
0.17
0.00
0.12
0.05
0.00
0.00
18.14
14.82
Pbhρtt
––
–0.0089
1.00
0.10
0.00
0.16
0.08
0.00
0.00
14.29
13.97
Pbhw
g,in
––
–0.0089
1.00
0.11
0.00
0.11
0.10
0.00
0.00
14.65
10.58
Pbhw
out
––
–0.0089
1.00
0.10
0.00
0.12
0.05
0.00
0.00
14.11
13.74
PttPbh
––
–0.0089
1.00
0.10
0.00
0.12
0.05
0.00
0.00
9.44
14.10
Pttρtt
––
–0.0028
1.00
0.30
0.00
0.16
0.08
0.00
0.00
24.02
21.89
Pttw
g,in
––
–0.0028
1.00
0.37
0.00
0.11
0.11
0.00
0.00
28.70
33.24
Pttw
out
––
–0.0028
1.00
0.28
0.00
0.13
0.05
0.00
0.00
22.61
20.94
woutw
g,in
––
–0.0024
1.00
0.43
0.00
0.12
0.11
0.00
0.00
30.75
37.15
66CONTROLLEDVARIABLESELECTION
Tab
le3.5:
Con
trollabilitydataforgas-lift
oilwellusingu1an
du2as
man
ipulatedvariab
les(M
IMO
controller)
CV
Pole
vector
MT
MK
SM
SG
MK
SG
d1
MK
SG
d2
MSG
d1
MSG
d2
γ1
γ2
γ3
Pabw
out
0.0034
1.00
0.12
0.00
0.08
0.03
0.00
0.00
7.55
13.20
15.31
Pabρtt
0.0034
1.50
0.14
2.00
0.11
0.05
0.34
0.12
10.39
12.16
16.53
Pabw
g,in
0.0034
1.00
0.16
0.00
0.09
0.07
0.00
0.00
11.43
10.98
12.40
PatPbh
0.0089
1.00
0.07
0.00
0.08
0.03
0.00
0.00
3.94
3.20
14.47
PatPtt
0.0031
1.59
0.13
11.00
0.08
0.03
0.96
0.62
7.36
7.76
8.16
Patw
out
0.0031
1.00
0.13
0.00
0.08
0.03
0.00
0.00
7.83
12.30
15.33
Patρtt
0.0031
1.52
0.15
1.87
0.11
0.05
0.32
0.11
11.23
12.89
16.74
Patw
g,in
0.0031
1.00
0.18
0.00
0.09
0.07
0.00
0.00
12.59
12.20
13.63
Pbhw
out
0.0089
1.00
0.07
0.00
0.08
0.04
0.00
0.00
3.15
13.30
75.80
Pbhρtt
0.0089
1.02
0.07
7.71
0.11
0.06
1.05
0.52
3.28
10.07
32.09
Pbhw
g,in
0.0089
1.00
0.07
0.00
0.09
0.09
0.00
0.00
3.35
3.20
3.58
PttPbh
0.0089
1.20
0.07
16.73
0.08
0.04
1.04
0.64
5.19
3.15
5.22
Pttw
out
0.0028
2.05
0.19
0.00
0.09
0.04
1.04
0.64
11.37
13.30
12.33
Pttρtt
0.0028
2.15
0.20
19.48
0.12
0.06
1.69
0.95
12.18
11.35
13.09
Pttw
g,in
0.0028
2.69
0.25
19.38
0.09
0.09
1.14
1.09
14.15
8.78
16.56
woutw
g,in
0.0024
1.00
0.30
0.00
0.10
0.09
0.00
0.00
13.30
11.32
21.55
3.3. Gas-Lifted Oil Well 67
0 5 10 15−1
0
1
u1
0 5 10 15−1
0
1
d1
d2
0 5 10 15−2
−1
0
1
2
wout
0 5 10 15−2
−1
0
1
2
setpoint P
bh
0 5 10 15−2
−1
0
1
2
time [h]
Pat
0 5 10 15−2
−1
0
1
2
time [h]
P
tt
Figure 3.13: H∞ control of gas-lift, CV = Pbh, MV = u1
tight control was required on the first controlled variable of the pair, andγ2 when tight control was on the second one in the pair.
Looking at cases with two controlled variables and u1 as the manipulatedvariable in Table 3.3, all cases including Pbh with tight control on Pbh resultin small γ values. However, there is no significant improvement in γ valuescompared to using the single controlled variable Pbh; simulation result ofcombing Pbh and wout with γ1 = 3.39 is shown in Figure 3.16. If Pbh is notavailable as directly measuremed either estimated, the next suitable CV iscombination of Pat and Ptt (two top-side pressures) with γ1 = 13.12. Thesimulation result for this case is given in Figure 3.17.
Table 3.4 shows controllability data for the gas-lift choke valve u2 as thesingle manipulated variables where relatively large peak of T and large γvalues for all CVs signals that u2 is not an effective manipulated variable.Combination of Pbh and Ptt gives the smallest γ value in Table 3.4. Sim-ulation results of the H∞ controller for this case with γ2 = 9.44 is shownin Figure 3.18. We conclude that the gas-lift choke valve is not a suitablemanipulated variable for the stabilizing control.
Using two manipulated variables, it is possible to have tight control onboth CVs when combiling two measurements; γ3 values in Table 3.5 werecalculated for this condition. γ1 and γ2 in Table 3.5 can be compared to
68 CONTROLLED VARIABLE SELECTION
0 5 10 15−1
0
1
u1
0 5 10 15−1
0
1
d1
d2
0 5 10 15−2
−1
0
1
2
wout
0 5 10 15−2
−1
0
1
2
Pbh
0 5 10 15−2
−1
0
1
2
time [h]
Pat
0 5 10 15−2
−1
0
1
2
time [h]
setpoint P
tt
Figure 3.14: H∞ control of gas-lift, CV = Ptt, MV = u1
0 5 10 15−1
0
1
u1
0 5 10 15−1
0
1
d1
d2
0 5 10 15−2
−1
0
1
2
wout
0 5 10 15−2
−1
0
1
2
P
bh
0 5 10 15−2
−1
0
1
2
time [h]
setpoint P
at
0 5 10 15−2
−1
0
1
2
time [h]
P
tt
Figure 3.15: H∞ control of gas-lift, CV = Pat, MV = u1
3.3. Gas-Lifted Oil Well 69
0 5 10 15−1
0
1
u1
0 5 10 15−1
0
1
d1
d2
0 5 10 15−2
−1
0
1
2
setpoint wout
0 5 10 15−2
−1
0
1
2
setpoint P
bh
0 5 10 15−2
−1
0
1
2
time [h]
Pat
0 5 10 15−2
−1
0
1
2
time [h]
P
tt
Figure 3.16: H∞ control of gas-lift, CV = [Pbh, wout], MV = u1
0 5 10 15−1
0
1
u1
0 5 10 15−1
0
1
d1
d2
0 5 10 15−2
−1
0
1
2
wout
0 5 10 15−2
−1
0
1
2
setpoint
0 5 10 15−2
−1
0
1
2
time [h]
setpoint P
at
0 5 10 15−2
−1
0
1
2
time [h]
setpoint P
tt
Figure 3.17: H∞ control of gas-lift, CV = [Pat, Ptt], MV = u1
70 CONTROLLED VARIABLE SELECTION
0 5 10 15−1
0
1
u2
0 5 10 15−1
0
1
d1
d2
0 5 10 15−2
−1
0
1
2
wout
0 5 10 15−2
−1
0
1
2
setpoint Pbh
0 5 10 15−2
−1
0
1
2
time [h]
Pat
0 5 10 15−2
−1
0
1
2
time [h]
setpoint Ptt
Figure 3.18: H∞ control of gas-lift, CV = [Pbh, Ptt], MV = u2
those in Table 3.3, but the cost function related to the H∞ problem forcalculating the γ3 values is different.
Looking at Table 3.5, the pairs with Pbh show small γ values, but com-pared to the γ values in Table 3.3, there is no substantial improvement. Thesimulation result of using the two top-side pressure measurement, Pat andPtt, using two manipulated variables is shown in Figure 3.19.
The pressures at top can be easily measured with good accuracy anda control structure using their combination (Figure 3.17) is recommended.However, by comparing simulation results in Figure 3.17 and Figure 3.19,one should notice that adding the secondary manipulated variable does notenhance the control performance.
Remarks
The choice of the suitable control structure is dependant on proper scalingof the controlled variables. For example for the gas-lift case, first we chosea small scaling factor for the mass flow rate and we wanted to control it ina tight bound. As a result, gain of the system with this output increasedand the control structures using the flow rate resulted in better performancecompared to those using the pressures.
In order to control the flow rate in a tight range, we must be ableto measure it accurately. However, this is unlikely for two-phase flow in
3.4. Summary 71
0 5 10 15−1
0
1
u1
u2
0 5 10 15−1
0
1
d1
d2
0 5 10 15−2
−1
0
1
2
wout
0 5 10 15−2
−1
0
1
2
P
bh
0 5 10 15−2
−1
0
1
2
time [h]
setpoint P
at
0 5 10 15−2
−1
0
1
2
time [h]
setpoint Ptt
Figure 3.19: H∞ control of gas-lift CV = [Pat, Ptt], MV = [u1, u2]
practice. Therefore, we chose a wider scaling factor for the flow rate. Onthe other hand, pressure can be measured more reliably, thus a small scalingfactor was used for pressures. Consequently, the control structures usingpressure measurements are shown to be superior for the gas-lift case study.
3.4 Summary
We perfomed the controllability analysis for two case studies, a well-pipeline-riser system and a gis-lifted oil well. Suitable CVs for stabilizng control wereidentified from the analysis. First, minimum achievable peaks of the differ-ent closed-loop transfer functions with each of the candidate CVs and theircombinations were calculated. Next, performance, robustness and inputusage requirements were integrated in a mixed-sensitivity control problem.From this, a new single measure (γ) was introduced to quantify the qualityof alternative control structures.
The bottom-hole pressure and the riser-base pressure are the best CVsfor SISO control of the well-pipeline-riser system. Because of the pressuredriven nature of the oulet flow in this case, the flow measurement shows alarger steady-state gain and consequently a better performance, comparedto the pipeline-riser case in [77]. Therefore, having an accurate measurementof the outlet flow rate of the choke valve, it can be used in a SISO controlscheme for stabilization.
72 CONTROLLED VARIABLE SELECTION
Combing one pressure measurement from the subsea with the outlet flowrate gives the best result for the well-pipeline-riser system. If the subseapressure measurements are not available, combining the top pressure andthe flow rate gives a satisfactory result too. If the flow measurement is notavailable, the top pressure combined with the density is able to stabilize thesystem in theory. However, the measurements Prt and ρrt and αl,rt are notsuitable CVs for a SISO control
For the gas-lifted oil well, the bottom-hole pressure is the best controlledvariable. Nevertheless, this variable often is not directly measurable. A con-trol structure using combination of two topside pressures from the annulusand the tubing was found to be effective to prevent the casing-heading in-stability. This can be implemented as a cascade control or the two topsidepressures can be used by an observer to estimate the state variables for astate feedback control.
The production choke valve is the main manipulated variable for thegas-lift well. However, we found that the adding the secondary manipulatedvariable (gas-lift choke) does not improve stabilization of the gas-lifted oilwells significantly.
Further, it was found that accuracy of the measureing devices (sensors)must be taken into account in order to scalde different outputs of a systemcorrectly.
Chapter 4
MANIPULATED
VARIABLE SELECTION
A top-side choke valve is usually used as the manipulated variable for anti-slug control of multi-phase risers at offshore oil-fields. With new advancesin the subsea technology, it is now possible to move top-side facilities to thesea floor. The two main contributions in this chapter are to consider analternative location for the control valve and to consider how to deal withnonlinearity. This research involved controllability analysis based on a sim-plified model fitted to experiments, simulations using the OLGA simulator,as well as an experimental study. It was concluded that a control valveclose to the riser-base is very suitable for anti-slug control, and its opera-tion range is the same as the top-side valve. However, a subsea choke valveplaced at the well-head can not be used for preventing the riser-slugging.The results provided in this chapter have been presented in [41].
4.1 Introduction
The oscillatory flow condition in offshore multi-phase pipelines is undesir-able and an effective solution is needed to suppress it [25]. Active controlof the topside choke valve is the recommended solution to maintain a non-oscillatory flow regime [71]. It also allows for larger valve openings andconsequently higher production rate [60], [62]. The control system used forthis purpose is called anti-slug control. This control system uses measure-ments such as pressure, flow rate or fluid density as the controlled variablesand a choke valve located at the top-side platform is the usual manipulatedvariable.
73
74 MANIPULATED VARIABLE SELECTION
By new advances in the mid-stream technologies, the subsea engineeringis an integral part of the oil production. The subsea separation and thesubsea compression are now standardized technologies used in practice, andmoving all the facilities to the sea floor is the ongoing trend. However,having all facilities at the subsea, the produced oil and gas are needed tobe transported to the sea level which involves using risers.
If we need to use the top-side choke valve for purposes other than thestabilizing control (e.g. safety and shut-down), a subsea solution for anti-slug control could be attractive. In order to explore possibilities of doinganti-slug control integrated with the subsea technology, we consider anti-slug control using subsea control valves in this chapter.
To use such a solution, we first need to consider if manipulating a sub-sea choke can prevent the riser slugging, and then where the control valvemust be located for an effective stabilizing control. Next, one must lookinto input-output pairing to choose the best controlled variable in terms ofrobustness and performance of the control loop.
We compare different manipulated variables (inputs, MVs) and con-trolled variables (outputs, CVs) in terms of robustness and performance forstabilizing control. This controllability analysis is done based on a simplifiedmodel of the system. We have extended the four-state simplified model in[34] to include two subsea choke valves, one at the wellhead and one close tothe riser-base. The simplified model is fitted to both the OLGA model andexperiments. Moreover, results from the controllability analysis are verifiedby simulations using the OLGA simulator as well as experiments.
The system is highly nonlinear and the gain of the system decreasesdrastically as we open the valve. In the closed-loop system we want tokeep the loop gain approximately constant for different operating points.Therefore, we need to increase the controller gain for large valve openings(lower pressure set-points).
4.2 Pipeline-Riser System
4.2.1 Experimental setup
The experiments were performed on a laboratory setup for anti-slug con-trol at the Chemical Engineering Department at NTNU. Fig. 4.1 shows aschematic presentation of the laboratory setup. The pipeline and riser inthe L-shaped setup are made from flexible pipes with 2 cm inner diameter.The length of the pipeline is 3 m, inclined downward with a 15◦ angle, andthe height of the riser is 3 m. A buffer tank for gas is used to simulate the
4.2. Pipeline-Riser System 75
Pump
Buffer
Tank
Water
Reservoir
Seperator
Air to atm.
Mixing Point
safety valve
P1
Pipeline
Riser
Subsea Valve
Top-side
Valve
Water Recycle
FT water
FT air
P3
P4
P2
Pin
Prb
Prt
Z2
Z1
Figure 4.1: Schematic diagram of experimental setup
effect of a long pipe with the same volume, such that the total resultinglength of the pipe would be about 70 m. There are two valves that may beused for control puposes; a topside valve and a subsea valve located closeto the riser base (see Fig. 4.1).
The feed into the pipeline is at constant flow rates; 4 litre/min of wa-ter and 4.5 litre/min of air. The separator pressure after the topside chokevalve is nominally constant at the atmospheric pressure. With these bound-ary conditions, the system switches from stable to unstable operation atZ1 = 15% opening of the top-side valve when the subsea choke valve isfully (Z2100%) open. To stabilize the system using manual choking of theriser-base subsea valve, we need to close this valve to less than Z2 = 8%.
4.2.2 OLGA model
First, we simulated the experimental rig in the OLGA simulator using thesame dimensions, and even including the buffer tank. The results for thisare not included, because we found them to be quite similar to our extendedOLGAmodel which includes an hypothetical oil well and its related wellheadvalve. Figure 4.2 shows a schematic of the final model that was used in thiswork. The oil well is vertical with height of 20 m, inner diameter of 0.02 m,and the reservoir pressure is fixed at 3.45 bar. We choose these parameterssuch that inflow conditions would be similar to the experimental setup. Theother dimensions and parameters are chosen very similar to the experimentalsetup.
In the final model with the oil well, we replaced the buffer tank by a220 m horizontal pipe, much longer than our estimate of 70 m, to get the
76 MANIPULATED VARIABLE SELECTION
same slug frequency as the experiments. In the experimental setup, thebuffer tank contains only the gas phase while in the OLGA model, similarto practical conditions, more than half of the pipeline volume is occupied byliquid. To adjust the slug frequency, the volume occupied by the gas phasein the pipeline is important. The period of oscillations was T = 68 sec.
PinPwh
Pbh
Prb
Prt
Z3Z2
Z1
Manipulated variables
Z3: wellhead valve
Z2: riser-base valve
Z1: topside valve
Controlled variablesPbh: bottom-hole pressure
Pwh: wellhead pressure
Pin: pipeline inlet pressure
Prb: riser base pressure
Prt: riser top pressure
Figure 4.2: OLGA case for well-pipeline-riser system
4.2.3 Simplified model
A four-state simplified model for severe-slugging flow in pipeline-riser sys-tems was presented in [34]. We have extended this model to include twosubsea valves; one at the well-head and one near the riser base. The oilwell dynamics are modelled by an additional state variable representing thetotal fluid mass in the well. The state variables of the augmented systemmodel are as these:
� mtw: mass of total fluid in well [kg]
� mgp: mass of gas in pipeline [kg]
� mlp: mass of liquid in pipeline [kg]
� mgr: mass of gas in riser [kg]
� mlr: mass of liquid in riser [kg]
4.2. Pipeline-Riser System 77
The state equations are the mass conservation laws,
mtw = wres − wwh (4.1a)
mgp = wg,in − wg,rb (4.1b)
mlp = wl,in − wl,rb (4.1c)
mgr = wg,rb − wg,out (4.1d)
mlr = wl,rb − wl,out (4.1e)
where,
� wres: mass flow from reservoir to well [kg/s]
� wwh: mass flow from wellhead to pipeline [kg/s]
� wg,rb: mass flow of gas at riser base [kg/s]
� wl,rb: mass flow of liquid at riser base [kg/s]
� wg,out: outlet gas mass flow [kg/s]
� wl,out: outlet liquid mass flow [kg/s]
These flow rates are calculated by valve type equations as given in [54].The simple model was fitted to the experiments by adjusting the followingsix parameters:
� Ka: correction factor for average gas fraction in well
� Kwh: wellhead choke valve constant
� Kh: correction factor for level of liquid in pipeline
� Kpc: production choke valve constant
� Kg: coefficient for gas flow through low point
� Kl: coefficient for liquid flow through low point
We refer to [34] and [54] for more details. The bifurcations diagrams, de-scribing the steady-state behaviour of the system as a function of the valveopening and the transition from stability to instability, are used to comparethe simplified model with experiments and the OLGA model [77]. Fig. 4.3and Fig. 4.4 show the bifurcation diagrams for the topside valve and the
78 MANIPULATED VARIABLE SELECTION
0 10 20 30 40 50 60 70 80 90 10010
20
30
40
50
Pressure at inlet of pipeline, Pin
Pin
[kpa
]
topside valve opening, Z1 [%]
experiment simple model olga model
0 10 20 30 40 50 60 70 80 90 1000
5
10
15
20
25
30
Pressure at top of riser, Prt
Prt [k
pa]
topside valve opening, Z1 [%]
experiment simple model olga model
Figure 4.3: Bifurcation diagrams for Z1 (two other valves fully open)
0 10 20 30 40 50 60 70 80 90 10010
20
30
40
50
60
Pressure at inlet of pipeline, Pin
Pin
[kpa
]
riser−base valve opening, Z2 [%]
experiment simple model olga model
0 10 20 30 40 50 60 70 80 90 1000
5
10
15
Pressure at top of riser, Prt
Prt [k
pa]
riser−base valve opening, Z2 [%]
experiment simple model olga model
Figure 4.4: Bifurcation diagrams for Z2 (two other valves fully open)
4.3. Controllability Analysis 79
subsea valve, respectively. The simplified model (thin solid lines) is com-pared to the experiments (red solid lines) and the OLGA model (dashedlines). In Fig. 4.3, the system has a stable (non-slug) flow when the topsidevalve opening Z1 is smaller than Z∗
1 = 15%, and it switches to slugging flowconditions for Z1 > 15%. We see three lines for slugging conditions. Theyare minimum and maximum pressure of the oscillations for slugging and thenon-slug flow pressure. The non-slug flow regime is unstable for Z1 > 15%,but it can be stabilized by using feedback control.
The corresponding bifurcation diagram for the riser-base valve is shownin Fig. 4.4. The OLGA model could not capture both steady-state and thecritical valve opening (Z∗
2 = 8%) at the same time for the riser-base valve; itwas only possible to get the critical valve opening correct by adjusting theCoefficient of Discharge of this valve. On the other hand, in the simplifiedmodel, we have more free parameters and we could fit the simplified modelcloser to the experiments (Fig. 4.4).
4.3 Controllability Analysis
A controllability analysis should reveal limitations on the achievable perfor-mance of a given input(s) and output(s) combination [74]. A controllabilityanalysis was used in Section 3.2 to find suitable controlled variables foranti-slug control when using the top-side choke valve as the manipulatedvariable. In this Section, we use a similar controllability analysis to com-pare the three alternative manipulated variables of the system (three controlvalves).
We compare minimum achievable peaks of three closed-loop transferfunctions (S, KS and SG), the output pole vectors and the steady-stategain as given in Tables 4.1, 4.2 and 4.3. It is desirable with a large steady-state gain |G(0)|, a large output pole vector and small values for the peaksof the closed-loop transfer function (MS , MKS and MSG).
The results for the top-side valve (Table 4.1) are similar to what was pre-sented in Section 3.2. The four subsea pressures are all suitable candidates,but the topside pressure (Prt) is not a good controlled variable because ofa large peak on S.
The controllability analysis for the riser-base subsea control valve (Ta-ble 4.2) shows that the pressure measurements upstream of this valve (Pbh, Pwh
and Pin) are good candidate controlled variables. The measurements down-stream this valve (Prb, Prt and wout) have small steady-state gains and arenot suitable.
The controllability analysis for the wellhead control valve (Table 4.3)
80 MANIPULATED VARIABLE SELECTION
Table 4.1: Controllability data for top-side choke valve Z1 = 40%
CV Value G(0)Polevector
MS MKS MSG
Pbh [kpa] 174.12 -1.26 12.09 1.00 2.69 0.00Pwh [kPa] 38.96 -1.48 13.17 1.00 2.47 0.00Pin [kpa] 24.02 -1.99 15.02 1.00 2.17 0.00Prb [kpa] 21.37 -2.08 23.41 1.00 1.39 0.00Prt [kpa] 3.07 -2.09 8.28 7.00 3.93 5.19wout [l/min] 4.17 0.08 13.47 1.00 2.42 0.00
Table 4.2: Controllability data for riser-base choke valve Z2 = 40%
CV Value G(0)Polevector
MS MKS MSG
Pbh [kpa] 164.26 -3.24 17.19 1.00 1.11 0.00Pwh [kpa] 28.86 -3.31 17.53 1.00 1.09 0.00Pin [kpa] 26.85 -3.47 18.30 1.00 1.05 0.00Prb [kpa] 22.82 0.18 28.66 1.30 0.67 0.94Prt [kpa] 2.25 0.02 2.08 1.00 9.23 0.00wout [l/min] 4.76 0.19 15.76 1.00 1.22 0.00
Table 4.3: Controllability data for well-head choke valve Z3 = 40%
CV Value G(0)Polevector
MS MKS MSG
Pbh [kpa] 170.37 -2.78 15.52 4.80 1.73 13.92Pwh [kpa] 27.67 -2.01 15.72 3.78 1.70 8.23Pin [kpa] 26.37 0.49 16.26 1.00 1.65 0.00Prb [kpa] 20.46 0.03 29.35 1.01 0.91 0.06Prt [kpa] 2.30 0.02 1.65 1.60 16.28 0.07wout [l/min] 4.40 0.17 11.17 1.81 2.40 0.64
4.4. Experimental Results 81
Table 4.4: Proportional gains used for different pressure set-points
Experiments OLGA Simulations
Pset [kPa] Kc Pset [kPa] Kc
25.0 15 24.8 0.224.0 20 23.9 0.423.5 25 23.2 0.623.0 30 22.7 0.822.5 40 22.5 1.022.0 50 22.2 1.221.7 60 22.0 1.421.5 70 21.8 1.821.3 8021.0 90
shows that none of the candidate controlled variables are very promising.The two pressure measurements upstream this valve have good steady-stategains, but have very large peaks for the sensitivity function. Among thefour measurements downstream of this valve, the best is the pressure atinlet of the pipeline (Pin). It has relatively high steady-state gain and thepeaks of the sensitivities are not large. We will therefore investigate thiscandidate controlled variable further in the simulations.
4.4 Experimental Results
The middle line in the bifurcations diagrams (Figure 4.3 and Figure 4.4)represents the desired non-slug flow. The slope of this line ( ∂y∂u = ∂P
∂Z ) rep-resents the process gain, and we note that the gain decreases in magnitudeand approaches zero as the valve opening increases. In order to stabilizethe system with larger valve openings (lower pressure set-points) we needto increase the controller gain. We used a simple PI controller implementedin LabView in the experiments. The integral time (Ti = 120 s) was keptconstant in the two experiments, and values of Kc for different pressureset-pints are given in Table 4.4. These values for the proportional gain werefound by trial and error. Developing a procedure for tuning the controllerparameters considering nonlinearity of the system is subject of the nextchapter.
The same set of set-point dependant controller gains (Table 4.4) wereused for all experiments, both with the topside and the subsea riser-basevalve. The results are shown in Figure 4.5 and Figure 4.6, respectively. For
82 MANIPULATED VARIABLE SELECTION
0 5 10 15 20 25 3020
22
24
26
time [min]
Pin
[kpa
]
Pressure at inlet of pipeline (Pin
)
set−pointmeasurement
0 5 10 15 20 25 300
50
100
Opening of topside choke valve (Z1)
Z1 [%
]
time [min]
Figure 4.5: Experiment with MV = Z1 and CV = Pin
the both valve locations we could stabilize the system without saturatingthe valve down to a set-point of 21.5 kPa. The average valve opening withthis set-point is 63% for the top-side valve and 59% for the subsea valve.
4.5 OLGA Simulation Results
As for the experiments, a simple PI controller was used in the OLGA simu-lations with tunings obtained by trial and error. Tuning values for differentpressure set-points are given in Table 4.4. Simulation results of controlusing the top-side choke valve (Figure 4.7) are shown in Figure 4.8 andthe results using the riser-base control valve (Figure 4.9) are shown in Fig-ure 4.10. The maximum achievable valve opening can be used as a measureof the benefit achieved from the anti-slug control. However, if two controlvalves have different sizes or Coefficients of Discharge, they produce differ-ent amounts of pressure drop for the same valve opening. Therefore, theminimum pressure set-point is used as a better measure, as it was used alsoby [62], especially when comparing two different valves. As shown in Fig-ure 4.8 and Figure 4.10, in the OLGA Simulations, both the control valvescan stabilize the system down to the same pressure set-point of 22 kPa. Thevalve opening with this set-point for the top-side valve is 42%, and for thesubsea valve, it is 21%.
4.5. OLGA Simulation Results 83
0 5 10 15 20 25 3020
22
24
26
time [min]
Pin
[kpa
]
Pressure at inlet of pipeline (Pin
)
set−pointmeasurement
0 5 10 15 20 25 300
50
100
Opening of valve at riser base (Z2)
Z2 [%
]
time [min]
Figure 4.6: Experiment with MV = Z2 and CV = Pin
Pin
Z3Z2
Z1
Figure 4.7: Control structure with MV = Z1 and CV = Pin
In the OLGA simulations in Figure 4.8 and Figure 4.10, we use anoil well as the boundary condition such that the inflow rates are pressuredriven. Indeed, one can notice from Figure 4.8 and Figure 4.10 that whenwe decrease the pressure set-point, the inlet mass flow rate from the oilwell increases. Thus, in addition to the pressure set-point, we can see thebenefit of the stabilizing control by looking at the inflow rates. The final
84 MANIPULATED VARIABLE SELECTION
0 5 10 15 2021
22
23
24
25
time [min]
Pin
[kpa
]
Pressure at inlet of pipeline (Pin
)
set−pointmeasurement
0 5 10 15 200
50
100
Opening of topside choke valve (Z1)
Z1 [%
]
time [min]
0 5 10 15 204
4.5
5
Inlet mass flow rate (win
)
win
[kg/
min
]
time [min]
Figure 4.8: OLGA simulation with MV = Z1 and CV = Pin
Pin
Z3Z2
Z1
Figure 4.9: Control structure with MV = Z2 and CV = Pin
production rate achieved by using both the control valves is the same valueof 4.4 kg/min.
Next, we consider using the wellhead valve (Figure 4.11). From the con-
4.5. OLGA Simulation Results 85
0 5 10 15 2021
22
23
24
25
time [min]
Pin
[kpa
]
Pressure at inlet of pipeline (Pin
)
set−pointmeasurement
0 5 10 15 200
50
100
Opening of valve at riser base (Z2)
Z2 [%
]
time [min]
0 5 10 15 204
4.5
5
Inlet mass flow rate (win
)
win
[kg/
min
]
time [min]
Figure 4.10: OLGA simulation with MV = Z2 and CV = Pin
trollability analysis in the previous section, we predicted that control of thebottom-hole pressure (Pbh) and well-head pressure (Pwh) using the valveat the well-head (Z3) is difficult because of large peaks of the sensitivitytransfer function. Indeed, it was not possible to control these two variablesby manipulating the wellhead valve in the OLGA simulations. On the otherhand, the steady-state gain of the inlet pressure (Pin) is larger than forthe other measurements downstream of the wellhead and it does not showany high peak. The simulation results in Figure 4.12 and Figure 4.13 showthat the inlet pressure can be regulated, but it causes the valve to close,thus it shuts down the production. In Figure 4.12, we increase the set-point, but it does not have much effect and flow rate decreases again aftera while. Because of small steady-state gain, by increasing the set-point thevalve opening does not change considerably. We decrease the set-point inFigure 4.13. Since the lower set-point is infeasible, the valve closes com-pletely but the pressure can not track the given set-point. In summary, thewellhead valve cannot be used for anti-slug control.
86 MANIPULATED VARIABLE SELECTION
Pin
Z3Z2
Z1
Figure 4.11: Control structure with MV = Z3 and CV = Pin
4.6 Discussion
In addition to the controllability analysis, there are some physical reasons forthe wellhead valve could not be used to stabilize the system. One intuitivereason is that the riser-slugging instability is because of dynamics betweenthe pipeline (volume of gas) and the riser (weight of liquid). Therefore,manipulating a valve at the inlet does not have any effect on this and cannotstabilize the unstable dynamics.
Considering the simulations in Fig. 4.12 and Fig. 4.13 using the wellheadvalve, the valve tends to close down. The reason for this is that the closed-loop system is internally unstable. In this situation, the riser becomes full ofthe liquid and only some gas bubbles will go out of the riser. The weight ofa column of the liquid in the riser maintains the pressure regulation withouta considerable flow out of the riser.
4.7 Summary
There is a good agreement between the OLGA model, simplified model andthe experiments in bifurcations diagrams for the topside valve. We could fitthe simple model closer to the experiments compared to the OLGA modelfor the subsea valve. Furthermore, the controllability analysis results basedon the simplified model are consistent with the simulations and experiments.
The control valve close to the riser-base is very suitable for anti-slugcontrol, and the resulted benefit in terms of the production rate is same
4.7. Summary 87
0 50 100 150 20021
22
23
24
25
time [min]
Pin
[kpa
]
Pressure at inlet of pipeline (Pin
)
set−pointmeasurement
0 50 100 150 2000
5
10
Opening of valve at well−head (Z3)
Z3 [%
]
time [min]
0 50 100 150 2000
0.5
1
1.5
2
Inlet mass flow rate (win
)
win
[kg/
min
]
time [min]
Figure 4.12: OLGA simulation with MV = Z3 and CV = Pin
0 50 100 150 20021
22
23
24
25
time [min]
Pin
[kpa
]
Pressure at inlet of pipeline (Pin
)
set−pointmeasurement
0 50 100 150 2000
2
4
6
8
10
Opening of valve at well−head (Z3)
Z 3 [%]
time [min]
0 50 100 150 2000
0.5
1
1.5
2
Inlet mass flow rate (win
)
win
[kg/
min
]
time [min]
Figure 4.13: OLGA simulation with MV = Z3 and CV = Pin
88 MANIPULATED VARIABLE SELECTION
as using the top-side valve. However, a subsea choke valve placed at thewellhead can not be used for preventing the riser-slugging. This valve closesdown and decreases the production rate drastically.
We could stabilize the system in the OLGA simulations and experimentsup to very large valve openings by considering nonlinearity of the systemand gain-scheduling of the proportional gain for a PI controller.
Chapter 5
LINEAR CONTROL
SOLUTIONS
The anti-slug control requires operation around an open-loop unstable op-erating point. One solution is to use a robust controller based on a mecha-nistic model, for this purpose, we consider H∞ control. As an alternative,we design an IMC (Internal Model Control) controller from a model iden-tified by a closed-loop step test. We obtain a second order IMC controllerthat can be implemented as a PID controller with a low-pass filter on itsderivative action. As the third solution, we consider using PI-control, whichis the preferred choice in the industry. However, appropriate tuning is re-quired for robustness against plant changes and large inflow disturbances.We obtain the PI-controler tuning from asymptotes of the proposed IMCcontroller. The proposed model identification and control solutions wereverified experimentally on two different test rigs. Furthermore, we showedthat robustness and performance of the IMC-PID can match a H∞ con-troller. However, the prosed IMC-PID is easier to tune compared to H∞
control. The results provided in this chapter are submitted for presentationin [35].
5.1 Robust Control Based on Mechanistic Model
Here, we linearize the four-state mechanistic model presented in Section 2.2by calculating Jacobian matrices. Then, we design the H∞ controller basedon the linear model. We use two approaches to obtain the robust controller,H∞ mixed-sensitivity design and H∞ loop-shaping design.
89
90 LINEAR CONTROL SOLUTIONS
y1
y2
K(s)
G(s)
WP
Wu
WTy
GENERALIZED PLANT P(s)
eu1
u2
+_
CONTROLLER
u
Figure 5.1: Closed-loop system for mixed sensitivity control design
5.1.1 H∞ mixed-sensitivity design
We consider an H∞ problem where we want to bound σ(S) for performance,σ(T ) for robustness and low sensitivity to noise, and σ(KS) to penalize largeinputs. These requirements may be combined into a stacked H∞ problem[74].
minK
‖N(K)‖∞, N
∆=
WuKSWTTWPS
(5.1)
where Wu, WT and WP determine the desired shapes of KS, T and S,respectively. Typically, W−1
P is chosen to be small at low frequencies toachieve good disturbance attenuation (i.e., performance), andW−1
T is chosento be small outside the control bandwidth, which helps to ensure goodstability margin (i.e., robustness). Wu is often chosen as a constant. Thesolution to this optimization problem gives a stabilizing controller K thatsatisfies [16], [23]:
σ(KS(jω)) ≤ γσ(W−1u (jω))
σ(T (jω)) ≤ γσ(W−1T (jω))
σ(S(jω)) ≤ γσ(W−1P (jω))
(5.2)
We assume four outputs for the model (Pin, Prt, wout and Qout); all the fouroutputs are in the y1 part and the particular output for feedback Pin is in they2 part of the generalized plant in Figure 5.1. The value of γ in equation(5.2) should be as small as possible for good controllability. However, itdepends on the design specifications Wu, WT and WP .
5.1. Robust Control Based on Mechanistic Model 91
N M
N M-1
K
u y
+
+
_
+
Figure 5.2: H∞ robust stabilization problem
5.1.2 H∞ loop-shaping design
We consider the stabilization of the plant G which has a normalized leftcoprime factorization
G = M−1N (5.3)
where we have dropped the subscripts from M and N for simplicity. Aperturbed plant model Gp can then be written as
Gp = (M +∆M )−1(N +∆N ) (5.4)
where ∆M and ∆N are stable unknown transfer functions which representthe uncertainty in the nominal plant model G. The objective of robuststabilization is to stabilize not only the nominal model G, but a family ofperturbed plants defined by
Gp ={
(M +∆M )−1(N +∆N ) : ‖[∆N ∆M ]‖∞ < ǫ}
(5.5)
where ǫ > 0 is then the stability margin [74]. To maximize this stabilitymargin is the problem of robust stabilization of normalized coprime factorplant description as introduced and solved by Glover and McFarlane [24].
For the perturbed feedback system of Figure 5.2, the stability propertyis robust if and only if the nominal feedback system is stable and
γK ,
∥
∥
∥
∥
[
KI
]
(I −GK)−1M−1
∥
∥
∥
∥
∞
≤ 1
ǫ(5.6)
Notice that γK is the H∞ norm from φ to
[
uy
]
and (I − GK)−1 is the
sensivity function for this positive feedback arragmenet. A samll γK iscorresponding to a large stability margin.
92 LINEAR CONTROL SOLUTIONS
s
! $
s
!
K
s
"
( )G s' u
Static Nonlinearity Linear time!invariant
u y u
u
( )G s
Figure 5.3: Block diagram for Hammerstein model
5.2 IMC Based On Identified Model
5.2.1 Model Identification
We use a Hammerstein model structure (Figure 5.3) to describe the desiredunstable operating point (flow regime). The Hammerstein model consistsof a series connection of a static nonlinearity (gain K) and a linear time-invariant dynamic system, G′(s). For identification of the unstable dynam-ics, we need to assume a structure. We first considered a simple unstablefirst-order plus delay model:
G(s) =Ke−θs
τs− 1=
be−θs
s− a(5.7)
where a > 0. If we control this system with a proportional controller withgain Kc0 (Figure 5.4), the closed-loop transfer function from the set-point(ys) to the output (y) becomes
y(s)
ys(s)=
Kc0G(s)
1 +Kc0G(s)=
Kc0be−θs
s− a+Kc0be−θs. (5.8)
In order to get a stable closed-loop system, we need Kc0b > a. The steady-state gain of the closed-loop transfer function is then
∆y∞∆ys
=Kc0b
Kc0b− a> 1. (5.9)
However, the closed-loop experimental step response (see Figure 5.5) showsthat the steady-state gain is smaller than one. Therefore, the model formin (5.7) is not a correct choice.
If we linearize the four-state mechanistic model in Section 2.2 aroundthe desired unstable operating point, we get a fourth-order linear model inthe form
G(s) =θ1(s+ θ2)(s+ θ3)
(s2 − θ4s+ θ5)(s2 + θ6s+ θ7). (5.10)
5.2. IMC Based On Identified Model 93
B
B
G(s) y(s) u e ys(s)
+ _ 0cK
Figure 5.4: Closed-loop system with conventional feedback. In the exper-imental step test we use C(s) = Kc0
.
This model contains two unstable poles, two stable poles and two zeros.Seven parameters (θi) must be estimated to identify this model. However, ifwe look at the Hankel Singular Values of the fourth-order model (Figure 5.6),we find that the stable part of the system has little dynamic contribution.This suggests that a model with two unstable poles is sufficient for controldesign. Using a model truncation (square root method), we obtained areduced-order model in the form of
G(s) =b1s+ b0
s2 − a1s+ a0, (5.11)
where a0 > 0 and a1 > 0. The model has two unstable poles and fourparameters, b1, b0, a1 and a0, need to be estimated. If we control theunstable process in (5.11) using a proportional controller with gain Kc0, theclosed-loop transfer function from set-point (ys) to output (y) becomes
y(s)
ys(s)=
Kc0(b1s+ b0)
s2 + (−a1 +Kc0b1)s+ (a0 +Kc0b0). (5.12)
This can be rewritten to the model used in [83]:
y(s)
ys(s)=
K2(1 + τzs)
τ2s2 + 2ζτs+ 1(5.13)
To estimate the four parameters (K2, τz, τ and ζ) in (5.13), we use six data(∆yp, ∆yu, ∆y∞, ∆ys, tp and tu) observed from the closed-loop response(see Figure 5.5). Then, we back-calculate the parameters of the open-loopunstable model in (5.11). Details are given in B.
5.2.2 IMC design for unstable systems
The Internal Model Control (IMC) design procedure is summarized in [58].The block diagram of the IMC structure is shown in Figure 5.7. Here, G(s)
94 LINEAR CONTROL SOLUTIONS
20 40 60 80 100 120 14025.5
26
26.5
27
27.5
28
28.5
29
time [sec]
y [kpa]
output
set point
!yp
!y!y
u
8
tpt
!ys
u
Figure 5.5: Experimental closed-loop step response for system stabilizedwith proportional control
1 2 3 40
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Order
abs
Hankel Singular Values
HSV(Stable Part of G)HSV(Unstable Part of G)
Figure 5.6: Hankel Singular Values of fourth order model
5.2. IMC Based On Identified Model 95
is the nominal model which in general has some mismatch with the realplant Gp(s). Q(s) is the inverse of the minimum phase part of G(s) andf(s) is a low-pass filter for robustness of the closed-loop system.
The IMC configuration in Figure 5.7 cannot be used directly for unstablesystems; instead we use the conventional feedback structure in Figure 5.4with the stabilizing controller
C(s) =Q(s)f(s)
1−G(s)Q(s)f(s). (5.14)
For internal stability, Qf and (1 − GQf) have to be stable. We use theidentified model from the previous section as the plant model:
G(s) =b1s+ b0
s2 − a1s+ a0=
k′(s+ ϕ)
(s− π1)(s − π2)(5.15)
and we get
Q(s) =(1/k′)(s− π1)(s − π2)
s+ ϕ(5.16)
We design the filter f(s) as explained in [58]:k = number of RHP poles + 1 = 3m = max(number of zeros of Q(s) - number of pole of Q(s) ,1) = 1 (tomake Q = Qf proper)n = m + k -1 = 3 (filter order)
The filter is in the following from:
f(s) =α2s
2 + α1s+ α0
(λs + 1)3, (5.17)
where λ is an adjustable closed-loop time-constant. We choose α0 = 1 toget integral action and the coefficients α1 and α2 are calculated by solvingthe following system of linear equations:
(
π12 π1 1
π22 π2 1
)
α2
α1
α0
=
(
(λπ1 + 1)3
(λπ2 + 1)3
)
(5.18)
Finally, from (5.14) the feedback version of the IMC controller becomes
C(s) =[ 1k′λ3 ](α2s
2 + α1s+ 1)
s(s+ ϕ). (5.19)
96 LINEAR CONTROL SOLUTIONS
f
Plant y e
+ _
( )f s ( )Q s u
( )G s _ + "
Figure 5.7: Block diagram of Internal Model Control system
5.2.3 PID implementation of IMC controller
Here, we obtain a PIDF tuning from the proposed IMC controller. TheIMC controller in (5.19) is a second order transfer function which can bewritten in form of a PID controller with a low-pass filter.
KPID(s) = Kc +Ki
s+
Kds
Tfs+ 1(5.20)
whereTf = 1/ϕ (5.21)
Ki =Tf
k′λ3(5.22)
Kc = Kiα1 −KiTf (5.23)
Kd = Kiα2 −KcTf (5.24)
For the controller work in practice, we require that Kc < 0 and Kd < 0;and we must choose λ such that these two conditions are satisfied. This wasobserved in the experiments.
5.3 PI Control
Next, we consider PI control. There are many approaches to get tuningvalues for PI control. For example, relay-feedback auto-tuning has beenused in [61] for PI tuning based on a first-order unstable model. Here, weobtain the PI tuning based on the IMC controller from the previous Section.We consider a PI controller in the following form
KPI(s) = Kc
(
1 +1
τIs
)
, (5.25)
The PIDF controller in (5.20) can be approximated by a PI-controller byconsidering the high- and low-frequency asymptotes of C(s) in (5.19).
Kc = lims→∞
C(s) =α2
k′λ3(5.26)
5.4. Small-Scale Experiments 97
τI =Kc
lims→0
sC(s)= α2ϕ (5.27)
5.4 Small-Scale Experiments
5.4.1 Experimental setup
Here, we use the small-scale experimental set-up introduced in Section 4.2.1to test the proposed PI and PID tuning rules.
5.4.2 Experiment 1: IMC-PID controller at Z=20%
The flow regime switches to slugging flow at a valve opening of Z = 15%,hence it is unstable at Z = 20%. We closed the loop with a proportionalcontroller with Kc0 = −10, and changed the set-point by 2 kPa (Figure 5.8).Since the response is noisy, a low-pass filter was used to reduce the noiseeffect. Then, we use the method described in Section 5.2.1 to identify theclosed-loop stable transfer function:
y(s)
ys(s)=
2.317s + 0.8241
19.91s2 + 2.279s + 1(5.28)
The identified closed-loop transfer function is shown by the red line in Fig-ure 5.8. From this, we back-calculate to an open-loop unstable processmodel:
G(s) =−0.012s − 0.0041
s2 − 0.0019s + 0.0088(5.29)
We select λ = 10 s for an IMC design to get the controller:
C(s) =−25.94(s2 + 0.07s + 0.0033)
s(s+ 0.35)(5.30)
The corresponding PIDF tuning values, as given in Section 5.2.3, are Kc =−4.44, Ki = −0.24, Kd = −60.49 and Tf = 2.81 s. Figure 5.9 shows thePIDF controller performance. This controller was tuned for Z = 20% valveopening, but it can stabilize the system up to Z = 32% valve opening whichshows good gain margin. In addition, we tested its delay margin by addingtime delay to the measurement. It was stable with 3 sec added time delay.
5.4.3 Experiment 2: PI tuning at Z=20%
Next, we obtain the PI tuning from the IMC controller (5.30) as explainedin Section 5.3. The PI tuning parameters are Kc = −25.95 and τI =
98 LINEAR CONTROL SOLUTIONS
0 20 40 60 80 100 120 14025
26
27
28
29
Pin
[kp
a]
t [sec]
data set−point filtered identified
Figure 5.8: Closed-loop step test for experiment 1
107.38 s. Figure 5.10 shows result of experiment using the PI controller.This controller was stable with 2 sec time delay.
5.4.4 Experiment 3: IMC-PID controller at Z=30%
We repeated the previous experiment at Z = 30% valve opening. We closedthe loop using a proportional controller with Kc0 = −20 and changed theset-point by 2 kPa (Figure 5.11). Then, we use the method explained inSection 5.2.1 to identify the closed-loop stable transfer function:
y(s)
ys(s)=
2.634s + 0.6635
13.39s2 + 2.097s + 1(5.31)
The identified closed-loop transfer function is shown by the red line in Fig-ure 5.11. Then, we back-calculate to an open-loop unstable system:
G(s) =−0.0098s − 0.0025
s2 − 0.0401s + 0.0251(5.32)
We select λ = 8 s for an IMC design to get the controller:
C(s) =−42.20(s2 + 0.052s + 0.0047)
s(s+ 0.251)(5.33)
The corresponding PIDF tuning parameters, as in Section 5.2.3, are Kc =−5.65, Ki = −0.79, Kd = −145.15 and Tf = 3.97 s. Figure 5.12 showsthe closed-loop performance of the PIDF controller. This controller wastuned for a Z = 30% valve openings, but it can stabilize the system up toa valve opening of Z = 50% which shows that controller has a good gainmargin. In addition, we tested its delay margin by adding time delay to themeasurement. It was stable with 2 sec added time delay.
5.4. Small-Scale Experiments 99
0 2 4 6 8 10 12 14 16 18 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 5.9: Result of IMC-PID controller for experiment 1
0 2 4 6 8 10 12 14 16 18 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 5.10: Result of PI controller for experiment 2
100 LINEAR CONTROL SOLUTIONS
0 20 40 60 80 100 120 14023
24
25
26
27
Pin
[kp
a]
t [sec]
data set−point filtered identified
Figure 5.11: Closed-loop step test for experiment 3
5.4.5 Experiment 4: PI tuning at Z=30%
The corresponding PI tuning obtained from the IMC controller in (5.33),as in Section 5.3, are Kc = −42.20 and τI = 53.53 s. Figure 5.13 shows anexperimental run using the PI controller. This controller was only stablewith added time delay less than 1 sec.
5.4.6 Experiment 5: H∞ mixed-sensitivity controller at Z=30%
We design the robust controller using the H∞ mixed-sensitivity method atthe operating point with Z = 30% with the following design specifications:
WT (s) =(s+ 1)2
(0.1s + 1)2, (5.34)
Wu = 1/100, (5.35)
WP0 = 1/3, (5.36)
WP (s) =s/Ms + ωB
s+ ωBA, (5.37)
where Ms = 3, ωB = 0.5 and A = 0.01. One should notice that WP0
does not apply any integral action and it is considered for the uncontrolledoutputs (Prt, wout and Qout) in the y1 part of the generalized plant. WP (s)applies an integral action with the bandwidth ωB = 0.5 on the controlledoutput Pin. With these specifications, γ = 3.86 is achieved, and we get a6th-order stabilizing controller. After removing two unimportant states, wehave
C(s) =−110.13(s + 4.58)(s2 + 0.025s + 0.0058)
s(s+ 0.22)(s2 + 6.71s + 15.03). (5.38)
5.4. Small-Scale Experiments 101
0 2 4 6 8 10 12 14 16 18 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
Controller OffController On
Controller Off
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 5.12: Result of IMC-PID controller for experiment 3
0 2 4 6 8 10 12 14 16 18 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
Controller OffController On
Controller Off
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 5.13: Result of PI controller for experiment 4
102 LINEAR CONTROL SOLUTIONS
Table 5.1: Comparison between IMC-PID controller and H∞ controllers
IMC-PIDH∞ mixed-sensitivity
H∞
loop-shaping
Gain Margin 0.38 0.46 0.29GM Frequency 0.19 0.18 0.16Phase Margin 61.68 50.63 64.44PM Frequency 0.52 0.43 0.64Delay Margin 2.07 2.07 1.77
‖S‖∞ 1.02 1.17 1.14‖T‖∞ 1.63 1.86 1.41‖KS‖∞ 43.17 36.86 55.67Zmax 50% 50% 58%
Figure 5.14 compares the Bode plot of this controller with the IMC-PIDcontroller in (5.33), and Figure 5.15 compares the closed-loop transfer func-tions of the controllers. Figure 5.16 shows experimental result of using theH∞-controller in (5.38); similar to the IMC-PID controller in (5.33), thisH∞ controller could stabilize the system up to a valve opening of Z = 50%,and it was stable with 2 sec added time delay. The margins of the twocontrollers are compared in Table 5.1. The gain-margin of the IMC-PID isbetter than that of the the H∞ controller.
It must be noted that although specifying suitable weights for the H∞ ismuch more complicated than tuning the IMC-PID using the filter time con-stant λ, the two controllers can achieve similar performance and robustness.
5.4.7 Experiment 6: H∞ loop-shaping controller at Z=30%
We use the IMC-PID controller in (5.33) to get a initially shaped plant forthe loop-shaping design. The code listed in Table 9.2 in [74] is used forthis design procedure. The lowest achievable value of γK in (5.6) for thiscase is 1.77. The procedure gives a eight-state stabilizing controller. Afterremoving unimportant states, we obtain the following third-order controller.
C(s) =−257.77(s2 + 0.0046s + 0.0024)
s(s+ 0.26)(s + 4.9)(5.39)
The gain-margin, the phase-margin and peaks for the closed-transfer func-tions for this controller are compared with the two previous controller inTable 5.1. This controller is faster; it has a better gain-margin and slightlyless delay-margin. The larger value for ‖KS‖∞ indicates a more aggres-sive control signal, but it shows a smaller value for ‖T‖∞. The Bode plot
5.5. Medium-Scale Experiments 103
10−2
10−1
100
101
0
20
40
60
80
Mag
[−]
IMC H∞ Mixed−sensitivity H∞ Loop−shaping
10−2
10−1
100
101
50
100
150
200
250
Pha
se [d
eg]
ω [Rad/s]
IMC H∞ Mixed−sensitivity H∞ Loop−shaping
Figure 5.14: Bode plot of H∞ controllers compared to IMC controller
of the loop-shaping controller is compared with the two other controller inFigure 5.14, and the closed-loop transfer functions are compared in Fig-ure 5.15.
Figure 5.17 shows experimental result of using the H∞ loop-shapingcontroller. When the controller is switched on, it is very aggressive whichwas expected from the large ‖KS‖∞. Because of the better gain-margin,it can stabilize the system in a wider range of the valve operation (from avalve opening of Z = 30% up to Z = 58%). We tested the delay margin byadding extra delay to the measured output; it was stable with 2 sec addeddelay.
5.5 Medium-Scale Experiments
5.5.1 Experimental setup
The tuning procedures were validated also on a medium-scale test rig. Thistest rig is an S-riser with a height of about 7 m. Other dimensions ofthis experimental set-up are shown in Figure 5.18. This riser is made fromstainless steel pipes with inner diameter of 50 mm. Similar to the small scalesetup, an air buffer thank is installed at inlet to emulate the effect of a long
104 LINEAR CONTROL SOLUTIONS
10−2
10−1
100
101
10−2
10−1
100
101
102
|S|
Sensitivity transfer function
IMC H∞ Mixed−sensitivity H∞ Loop−shaping
10−2
10−1
100
101
10−2
10−1
100
101
102
|T|
Complementary sensitivity transfer function
IMC H∞ Mixed−sensitivity H∞ Loop−shaping
10−2
10−1
100
101
10−2
100
102
104
Input usage
ω [Rad/s]
|KS|
IMC H∞ Mixed−sensitivity H∞ Loop−shaping
Figure 5.15: Closed-loop transfer functions for H∞ Controllers comparedto IMC controller
5.5. Medium-Scale Experiments 105
0 2 4 6 8 10 12 14 16 18 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 5.16: Result of mixed-sensitivity H∞ controller in experiment 5
0 5 10 15 20
15
20
25
30
35 open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 5 10 15 200
20
40
60
80
Controller OffController On
Controller Off
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 5.17: Result of loop-shaping H∞ controller in experiment 6
106 LINEAR CONTROL SOLUTIONS
7.04 2.83 2.63 1.41
13.96
4.01
2.70
6.43
Figure 5.18: Experimental setup for S-riser (all values in meter)
0 100 200 300 400 500 600145
150
155
160
165
Pin
[kp
a]
t [sec]
data set−point filtered identified
Figure 5.19: Closed-loop step test on medium-scale rig in experiment 7
pipeline with the same volume. The volume of the buffer tank is 200 litres;this is equivalent 101.86 m of pipe. The inlet flow rates to the system are0.0024 kg/sec air and 0.3927 kg/sec water. The outlet separator pressureis constant at the atmospheric pressure. With these boundary conditions,the system switches from non-slug to slugging flow conditions at Z∗ = 16%opening of the topside valve.
5.5.2 Experiment 7: IMC-PID controller at Z=18%
Figure 5.19 shows a closed-loop step test performed on the S-riser. A pro-portional controller with the gain Kc0 = −250 was used for the test. Thepressure set-point before the step test was 155 kPa which results in a valveopening of Z = 18% (region of unstable open-loop operation). We identified
5.5. Medium-Scale Experiments 107
0 5 10 15 20 25 30 35120
140
160
180
200
220
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 5 10 15 20 25 30 350
20
40
60
80
Controller Off Controller On
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 5.20: Experimental result of IMC-PID controller on medium-scalesetup in experiment 7
an unstable model as the following:
G(s) =−5.6 × 10−4(s+ 0.082)
s2 − 0.069s + 0.0040(5.40)
By choosing λ = 24.5, we designed the following IMC controller:
C(s) =−340.75(s2 + 0.0052s + 0.00036)
s(s + 0.0816)(5.41)
The corresponding PID tuning are Kc = −3.47, Ki = −1.49, Kd = −4.13×104 and Tf = 12.25. Experimental result of control using this PID tuningis shown in Figure 5.20. In this experiment, we decreased the set-pointuntil the system becomes unstable. This controller was able to control thesystem up to a Z = 32% valve opening, which is two time of the criticalvalve opening Z∗ = 16%.
5.5.3 Experiment 8: PI tuning at Z=18%
The PI tuning obtained from the IMC controller in (5.41) are Kc = −340.75and τI = 229.23. The experimental result is given in Figure 5.21 wheresystem was stabilized up to Z = 24%.
108 LINEAR CONTROL SOLUTIONS
0 5 10 15 20 25 30120
140
160
180
200
220
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 5 10 15 20 25 300
20
40
60
80
Controller Off Controller On
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 5.21: Experimental result of PI controller on medium-scale setupin experiment 8
5.6 Summary
A model structure including two unstable poles and one zero was used toidentify an unstable model for the desired non-slug flow dynamics which areopen-loop unstable. The model parameters were estimated from a closed-loop step test, and the identified model was used for an IMC design. Theproposed IMC controller can be implemented as a PID-F controller thatresults in good gain margin and phase margin. PI tunings were also ob-tained from the IMC controller. This scheme was tested experimentally ontwo experimental rigs (the small-scale and the medium-scale S-riser). Theperformance and the robustness of the IMC-PID controller was very similarto the mixed-sensitivity H∞ controller. The two controller can stabilize thesystem up to the same valve opening (Z = 50%), and they were stable with2 sec extra delay. However, compared to the H∞ controller, the IMC-PIDis easier to tune and it does not require a mechanistic model.
In addition to the H∞ mixed-sensitivity design, the loop-shaping designwas tested where the new IMC controller was used to obtain the initiallyshaped plant. The loop-shaping procedure improves the gain-margin. Thefinal controller is faster and generates an aggressive control action, but itcan stabilize the system up to a larger valve opening (Z = 58%).
Chapter 6
NONLINEAR CONTROL
SOLUTIONS
Instability and the inverse response behaviour make anti-slug control atoffshore oil-fields an interesting control problem where a robust solutionconsidering the nonlinearity of the system is required. Existing anti-slugcontrol systems become unstable after some time, because of inflow distur-bances or plant changes. The nonlinearity at different operating conditionsis one source of plant changes. In addition, the time delay is another prob-lematic factor for stabilization. The nonlinearity can be counteracted bymodel-based nonlinear controllers or by gain-scheduling of multiple linearcontrollers. A summary of the different anti-slug control solutions proposedin previous works is given in Table 6.1.
In this chapter, We test four new control solutions experimentally. First,we use state feedback based on state estimation using nonlinear observers.Second, we apply feedback linearization with measured outputs. Then, weconsider a PI controller with adaptive gain correction based on the static
Table 6.1: success of different control solutions in previous works
method \ measurementtopsidepressure
subseapressure
using bothpressures
top pressure& flow
PID/H∞/LQG [71], [75] † © © ©
Linear observer [37] † © − −Nonlinear observer [37] § † † −
Feedback linearization [38] − − © −Back stepping [48] − © − −−: not investigated ©: works well §: not robust †: doesn’t work
109
110 NONLINEAR CONTROL SOLUTIONS
gain of the system. Finally, we use gain scheduling for IMC (Internal ModelControl) based on linear models identified from closed-loop step tests. Wecompare these four approaches in terms of robustness and operation range,more precisely, we compare delay margin of the controllers and the maxi-mum achievable valve opening for each control solution. The results pro-vided in the chapter have been presented in [37], [38], [35], [36].
6.1 Pipeline-Riser System
6.1.1 Simplified dynamical model
A four-state simplified model for the severe-slugging flow was presented inSection 2.2. The state variables of this model are as:
� mgp: mass of gas in pipeline [kg]
� mlp: mass of liquid in pipeline [kg]
� mgr: mass of gas in riser [kg]
� mlr: mass of liquid in riser [kg]
The four state equations of the model are
mgp = wg,in − wg (6.1)
mlp = wl,in − wl (6.2)
mgr = wg − αw (6.3)
mlr = wl − (1− α)w (6.4)
Fig. 6.1 shows a schematic presentation of the model. The inflow rates ofgas and liquid to the system, wg,in and wl,in, are assumed to be constant.The flow rates of gas and liquid from the pipeline to the riser, wg andwl, are determined by pressure drop across the riser-base where they aredescribed by virtual valve equations. The outlet mixture flow rate, w, isdetermined by the opening percentage of the top-side choke valve, u, which isthe manipulated variable of the control system. The different flow rates andthe gas mass fraction, α, in the equations (6.1)-(6.4) are given by additionalmodel equations presented in Section 2.2.
6.1. Pipeline-Riser System 111
wg,in
wl,in
w
Ps
Lr
Prt
u
Prb
wg
wl
Pin
Figure 6.1: Schematic presentation of system
Pump
Buffer
Tank
Water
Reservoir
Seperator
Air to atm.
Mixing Point
safety valve
P1
Pipeline
Riser
Subsea Valve
Top-side
Valve
Water Recycle
FT water
FT air
P3
P4
P2
Figure 6.2: Schematic diagram of experimental setup
112 NONLINEAR CONTROL SOLUTIONS
6.1.2 Experimental setup
The experiments were performed on a laboratory setup for anti-slug con-trol at the Chemical Engineering Department of NTNU. Fig. 6.2 shows aschematic representation of the laboratory setup. The pipeline and the riserare made from flexible pipes with 2 cm inner diameter. The length of thepipeline is 4 m, and it is inclined with a 15◦ angle. The height of the riseris 3 m. A buffer tank is used to simulate the effect of a long pipe with thesame volume, such that the total resulting length of pipe would be about70 m.
The topside choke valve is used as the input for control. The separatorpressure after the topside choke valve is nominally constant at atmosphericpressure. The feed into the pipeline is assumed to be at constant flowrates, 4 litre/min of water and 4.5 litre/min of air. With these boundaryconditions, the critical valve opening where the system switches from stable(non-slug) to oscillatory (slug) flow is at Z∗ = 15% for the top-side valve.
Bifurcations diagrams, describing the steady-state and the dynamicsof this system, are used to fit the model to the experimental rig ([77]).Fig. 6.3 shows the bifurcation diagrams of the simplified model (solid lines)compared to the those of the experiments (dashed lines). The system hasa stable (non-slug) flow when the valve opening Z is smaller than 15%,and it switches to slugging flow conditions for larger valve openings. Theminimum and maximum of the oscillations of the slugging together with thesteady-state (in the middle) are shown in Fig. 6.3.
The desired steady-state (middle line) in slugging condition (Z > 15%)is unstable, but it can be stabilized by using control. The slope of the steady-state line is the static gain of the system, G = ∂y/∂u = ∂Pin/∂Z. As thevalve opening increase this slope decreases, and the gain finally approachesto zero. This makes control of the system with large valve openings verydifficult. The controller should keep the loop-gain (L = KG, where G is theprocess gain and K is the controller gain) constant in order to stabilize thesystem over the whole range of operation.
6.2 State Feedback with Nonlinear Observer
It has been shown that it is difficult to avoid slugging in pipeline-riser sys-tems when using only the top-side pressure measurement [77]. The reasonis that the unstable Right Half-Plane (RHP) zeros of the system with thismeasurement are relatively close to the unstable poles of the system, andconsequently the sensitivity transfer function of the system has an unavoid-
6.2. State Feedback with Nonlinear Observer 113
0 10 20 30 40 50 60 70 80 90 10010
20
30
40
50
Pressure at inlet of pipeline (Pin
)
Pin
[kpa
]
valve opening [%]
experiment simple model
0 10 20 30 40 50 60 70 80 90 1000
10
20
30
Pressure at top of riser (Prt)
Prt [k
pa]
valve opening [%]
experiment simple model
Figure 6.3: Bifurcation diagrams of simplified model (solid) compared toexperiments (dashed)
114 NONLINEAR CONTROL SOLUTIONS
able large peak. On the other hand, the pressure measurement at the subseais a suitable controlled variable, and a simple PI controller is used in practice[25].
If only the top-side pressure measurement is available, a conventionalcontrol solution is to design an observer which uses the top-side pressuremeasurement to estimate the states of the system including the bottompressure, and then use these estimates for control. However, the fundamen-tal controllability limitation of unstable zero associated with the top-sidepressure can not be bypassed by the observer. Nevertheless, we want to seeif this solution is applicable for anti-slug control, and which kind of observeris more suitable.
The Extended Kalman Filter has been previously used for estimatingthe bottom pressure and other other state variables in gas-lifted oil wells[20]. Moreover, a nonlinear observer based on back-stepping design wasproposed for this purpose [2], and it was tested in experiments [18]. Thegas-lift system has a similar unstable zero dynamic associated with thetubing pressure, but the tubing pressure in combination of the other top-side measurement has been used in [18], [20].
A nonlinear Luenberger-type observer that uses only the topside pres-sure to reproduce oscillations of the riser-slugging instability was proposedin [14], and then used for anti-slug control of a pipeline-riser system in ex-periments [56]. This observer was designed based on a three-state simplifiedmodel of the system [13]. It was found that in their case the peak of thesensitivity transfer function of the topside pressure was not large, and itdid not impose limitations on the controllability [12]. However, as discussedlater, this is true only for small valve openings, and we can not open thevalve for large production rates when using the top-side pressure.
We use the four-state simplified model presented in Section 2.2 for theobserver and control design in this work. The model is fitted to experimen-tal data by adjusting four parameters, and it shows good agreement withexperiments.
Three types of observer using the top-side pressure were tested experi-mentally. These are a Luenberger-type nonlinear high-gain observer, a stan-dard Unscented Kalman Filter (UKF), and a UKF modified to incorporatethe high-gain observer concept [51] which we call “Fast UKF”.
The term “High-Gain observer” can cause confusion in some cases. Onemay say a observer is not high-gain, because the observer gain is a smallnumber (e.g. K = 0.1) [56] [15]. However, it depends on which unit is usedfor the measurement; for example there is a scaling factor of 103 between Paand kPa for pressure. Hence, the speed (convergence rate) of the observer
6.2. State Feedback with Nonlinear Observer 115
Process
Observer
rtP
(Riser top pressure)
x rtP
+-
ssx u
ˆinP
+
++
-
-Kc
Ki r
Figure 6.4: Block diagram of closed-loop system
is a better indicator.As illustrated in Figure 6.4, we want to estimate the state variables of the
system by use of an observer and use the estimates for state feedback. Theseparation principle allows us to separate the design into two tasks. First, wedesign a state feedback controller that stabilizes the system and meets otherdesign specifications. Then an output feedback controller is obtained byreplacing the state x by its estimate x provided by observers [51]. However,the separation principle does not hold in general for nonlinear systems, andwe have to test this solution for the anti-slug control by experiments.
6.2.1 State feedback
As shown in Figure 6.4, we apply full state feedback by using the estimatedstates. In addition, to prevent drift from the operating point, integral actionis added by integrating the set-point deviation for the estimated subseapressure (Pin). The total control action can be expressed as
u(t) = −Kc(x(t)− xss) +Ki
t∫
0
(Pin(τ)− r)dτ . (6.5)
Here, Kc is a linear optimal controller calculated by solving Riccati equationand Ki is a relatively small integral gain.
6.2.2 Nonlinear Observers
Unscented Kalman Filter
First, we consider the standard form of the Unscented Kalman Filter (UKF)as explained by [47]. The nonlinear state space system is given as
xk = f(xk−1, uk−1) + vk−1, (6.6a)
116 NONLINEAR CONTROL SOLUTIONS
Table 6.2: Unscented Kalman Filter algorithm
(1) Prediction step:
Xk−1 = [xk−1 . . . xk−1] +√c[√
Pk−1 −√Pk−1
]
Xk = f(Xk−1, uk−1)
x−
k = XkWm
P−
k = XkWc[Xk]T +Qk−1
(2) Update step:
X−
k =[
x−
k . . . x−
k
]
+√c
[
√
P−
k −√
P−
k
]
Y −
k = h(X−
k )µk = Y −
k Wm
Sk = Y −
k Wm
[
Y −
k
]T+Rk
Ck = X−
k Wc
[
Y −
k
]T
(3) Compute filter gain Kk, updated state mean xk and covariance Pk:Kk = CkS
−1
k
xk = x−
k +Kk [yk − µk]Pk = P−
k −KkSkKTk
yk = h(xk, uk) + wk, (6.6b)
where vk−1 is a vector of Gaussian zero-mean process noise with the covari-ance matrix Qk−1, representing model error and disturbances. wk is a vectorof Gaussian zero-mean measurement noise with the covariance matrix Rk.
We assume that the state vector is a random variable with mean valuexk−1 and the covariance matrix Pk−1, undergoing the nonlinear transform off . The general problem is to find the statistics of this random variable afterthe nonlinear transform (xk). The main idea behind the UKF, unlike theMonte Carlo simulations which need a large number of samples, is that thestatistics of a nonlinear transformation can be found with fair accuracy byonly a limited number of samples called sigma points. The samples in MonteCarlo simulations are chosen randomly, whereas in the UKF algorithm thesigma points are chosen deterministically. Similar to the EKF, the UKFalgorithm has a predictor/corrector nature, but we do not need to linearizethe model by calculating the Jacobian matrices. The UKF algorithm issummarized in Table 6.2.
We do not discretize the model or the observers, instead we use Matlab(ode15s solver) to integrate the model from time k−1 to k. Because of smalldimensions of the laboratory set-up, the frequency of the unstable dynamicsare relatively high, and we need a small sampling time (Ts = 0.1 sec). Withthe four-state model, we get eight sigma points, and integrating the modeleight times needs more computation time than the sampling time. We solved
6.2. State Feedback with Nonlinear Observer 117
this problem by using parallel processing in Matlab.
High-Gain Luenberger observer
A high-gain observer, under certain conditions, guarantees that the outputfeedback controller recovers the performance of the state-feedback controllerwhen the observer gain is sufficiently high. The observer gain is designedso that the observer is robust to uncertainties in modeling the nonlinearfunctions. The structure of the high-gain observer is similar to the one usedin [14]:
˙z1 = f1(z)
˙z2 = f2(z) (6.7)
˙z3 = f3(z) +1
ǫ(y − y)
˙z4 = f4(z)
where
� z1, mass of gas in pipeline (mgp)
� z2, mass of liquid in pipeline (mlp)
� z3, pressure at top of riser (Prt)
� z4, mass of liquid in riser (mlr)
and 1ǫ is the high gain. Three of the equations (f1, f2 and f4) are the same
as the model in equations (6.1), (6.2) and (6.4) in Section 6.1. For thethird state equation (f3), we transform the state into top pressure which isa measurement (y = z3 and y = z3). We use z3 = Prt, because it is directlyrelated to the mass of gas in the riser (mgr) which is the third state of themodel:
Prt =mgrRTr
MG
(
Vr − mlg
ρl
) (6.8)
f3(z) =dPrt
dt(6.9)
We get the time derivative of the top-pressure by using partial derivatives:
dPrt
dt=
∂Prt
∂mgrmgr +
∂Prt
∂mlrmlr (6.10)
118 NONLINEAR CONTROL SOLUTIONS
where∂Prt
∂mgr=
a
b−mlr(6.11)
∂Prt
∂mlr=
amgr
(b−mlr)2(6.12)
In (6.11) and (6.12), a = RTrρl/MG and b = ρlVr are model constants; seeSection 2.2 for details of the model.
Fast UKF
One advantage of the Unscented kalman Filter (UKF) over a simple high-gain observer is that the UKF relates the measurements to all the stateequations with appropriate gains. Moreover, a Kalman Filter averages outmeasurement noise, and thus the estimates are less sensitive to noise. How-ever, the standard UKF is not sufficiently robust in closed-loop for ourapplication. In most of the experimental runs, the closed-loop system usingthe UKF was not able to stabilize the flow. However, by increasing theobserver gain, promising results were achieved. Although the observer gainfor the UKF can be increased by using a small value for Rk and a largevalue for Qk−1, there is a limitation with this approach. By looking at theUKF algorithm, the observer gain is
Kk =X−
k Wc
[
Y −
k
]T
Y −
k Wm
[
Y −
k
]T+Rk
. (6.13)
One notices that the observer gain is highly dependent on the scale of statesand measurements. In the case of measuring one of states, the maximumgain in direction of the measured state is 1. Even though the UKF, similarto the linear Kalman Filter, is optimal for estimation errors at steady-state,it can not guarantee robustness of the closed-loop system.
We aim to combine the advantages of the UKF with the robustnessproperties of a high-gain observer. We implement this idea in a basic andsimple way. We use the transformed model, similar to (6.7), instead of theoriginal model in (6.1)-(6.4). It is similar to the high-gain observer, butwithout the observer term.
z1 = f1(z)
z2 = f2(z) (6.14)
z3 = f3(z)
z4 = f4(z)
6.2. State Feedback with Nonlinear Observer 119
These are the model equations, only the third state has been transformed.We do not specify the observer term in the state equations explicitly, neitherdo we determine the observer gain directly; we let the Unscented Transfor-mation calculate the four elements of the observer gain. The matrix Qk
in the UKF algorithm represents process noises, and larger Qk leads to alarge observer gain. The third state of the model in (6.14) is measured. Bychoosing suitable values for the elements of Qk, we can put more weight inthe direction of the measured state. We construct Qk as follows:
Qk = diag(qmin, qmin, qmax, qmin) (6.15)
where qmin and qmax are treated as tuning parameters. The UKF algo-rithm calculates the observer gains in an adaptive manner, and unlike asimple high-gain observer, it gives four elements corresponding to the fourstates. As suggested in [6], we incorporated the innovation information(Mean Square Error in a moving window) to make Rk and Qk adaptive aswell. This idea was tested in simulations, using OLGA simulator as the realprocess, also in experiments. We found that the closed-loop system is morerobust in experiments when we have Rk and Qk constant.
Since we are using the UKF algorithm with a measured state (y = z3),the maximum observer gain is 1. To have better control of the observer gain,we scale the states and measurements of the system based on their steady-state values. All the states are normalized to 100, and the measurement isnormalized to 1000. This scaling implies that when the UKF gives a gain of1, we have enforced a factor of 10. However, we present the measurementswith their original scales in the experimental results.
6.2.3 Experimental results
Measuring top-side pressure
The experimental result using the high-gain Luenberger observer is shown inFigure 6.5. The same experiment using the Fast UKF is shown in Figure 6.6.Comparing Figures 6.5 and 6.6, we see that the Fast UKF behaves betterand, unlike the high-gain observer, does not show any oscillation when thecontroller is on.
As shown in Figure 6.3, the open-loop system switches from stable toslugging flow for valve openings Z > 15%. We were able to stabilize thesystem up to Z = 20% by using nonlinear observers when the top-sidepressure without any delay was measured. However, as the valve openingincreases, control becomes more difficult [40].
120 NONLINEAR CONTROL SOLUTIONS
The standard UKF was not able to stabilize the system for most ofexperimental runs, while the two other observers were successful in all ex-periment runs with Z = 20%. The result of using the standard UKF, whenthe system can be stabilized, is similar to Figure 6.6. We do not presentthe result here due to space limitation.
Measuring subsea pressure
Based on previous controllability analysis [77], [40], it is much easier tostabilize the system using the subsea pressure instead of the top-side pres-sure. Indeed, we found that the system could be stabilized up to Z = 40%valve opening by controlling subsea pressure using a simple PI control (Fig-ure 6.7), that is, with no observer.
Next, we considered state feedback control based on observers using thesubsea pressure as the only measurement. Surprisingly, it was not possi-ble to stabilize the system when used any of the fast observers (high-gainLuenberger and fast UKF). We constructed observers for the subsea pres-sure measurement, as done in the previous section, by transforming the firststate by choosing z1 = Pin.
Figure 6.8 shows the result of open-loop estimation by a Luenbergerobserver with a large gain (ǫ = 10−4), where the subsea pressure is themeasurement used by the observer. The top-side pressure and its relatedstate variables are not correctly estimated, and this seems to be the reasonwe were not able to stabilize the system. This happens when we increasethe gain of the observer for supposedly robustness of the closed-loop system.
On the other hand, we used a linear Kalman Filter with the subseapressure measurement, and it was possible to stabilize the system up toZ = 40% valve opening. As shown in Figure 6.9, an aggressive controlaction is needed to stabilize the system at this operating point.
We also performed closed-loop experiments where we started the experi-ment in closed-loop to keep the system always around its equilibrium point,and the results were the same; none of the three nonlinear observers workedproperly. The Fast UKF was able to stabilize the system, but it was notrobust against disturbance in boundary (inflow) conditions. We summarizeperformance of different observers for state estimation in Table 6.3, andperformance of different methods for stabilizing control in Table 6.4.
6.2. State Feedback with Nonlinear Observer 121
0 5 10 15 20 25 3015
20
25
30
35
40
t [min]
Pin
[kP
a]
inlet pressure (estimated by observer)
open−loop stable
open−loop unstable
set−pointactualobserver
0 5 10 15 20 25 300
5
10
15
t [min]
Prt [k
Pa]
topside pressure (measurement used by observer)
open−loop stable
open−loop unstable
actualobserver
0 5 10 15 20 25 300
20
40
60
80
100
ControllerOff
Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.5: Control experiment with top-side pressure (Prt) measurementused by high-gain observer, Z = 20%
122 NONLINEAR CONTROL SOLUTIONS
0 5 10 15 20 25 3015
20
25
30
35
40
t [min]
Pin
[kP
a]
inlet pressure (estimated by observer)
open−loop stable
open−loop unstable
set−pointactualobserver
0 5 10 15 20 25 300
5
10
15
t [min]
Prt [k
Pa]
topside pressure (measurement used by observer)
open−loop stable
open−loop unstable
actualobserver
0 5 10 15 20 25 300
20
40
60
80
100
ControllerOff
Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.6: Control experiment with top-side pressure (Prt) measurementused by Fast UKF, Z = 20%
6.2. State Feedback with Nonlinear Observer 123
0 5 10 15 20 25 3015
20
25
30
35
40
t [min]
Pin
[kP
a]
inlet pressure (controlled variable)
open−loop stable
open−loop unstable
set−pointmeasurement
0 5 10 15 20 25 300
5
10
15 open−loop stable
open−loop unstable
topside pressure
t [min]
Prt [k
Pa]
0 5 10 15 20 25 300
20
40
60
80
100
ControllerOff
Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.7: Control using subsea pressure measurement (Pin) and PI con-troller (Kc = −10, Ti = 200sec), Z = 40%
124 NONLINEAR CONTROL SOLUTIONS
0 1 2 3 4 5 6 7 8 9 1015
20
25
30
35
40
open−loop stable
open−loop unstable
t [min]
Pin
[kP
a]
inlet pressure (measurement used by observer)
actualobserver
0 1 2 3 4 5 6 7 8 9 100
5
10
15 open−loop stable
open−loop unstable
t [min]
Prt [k
Pa]
topside pressure (estimated by observer)
actualobserver
0 1 2 3 4 5 6 7 8 9 100
20
40
60
80
100
Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position
Figure 6.8: Open-loop estimation using High-Gain observer and measur-ing subsea pressure (Pin), Z = 20%
6.2. State Feedback with Nonlinear Observer 125
0 5 10 15 20 25 3010
15
20
25
30
35
40
t [min]
Pin
[kP
a]
inlet pressure (measurement used by observer)
open−loop stable
open−loop unstable
set−pointactualobserver
0 5 10 15 20 25 30−5
0
5
10
15
t [min]
Prt [k
Pa]
topside pressure (estimated by observer)
open−loop stable
open−loop unstable
actualobserver
0 5 10 15 20 25 300
20
40
60
80
100
ControllerOff
Controller OnController Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.9: Control experiment with subsea pressure measurement (Pin)used by linear KF, Z = 40%
126 NONLINEAR CONTROL SOLUTIONS
6.2.4 Discussion
Measuring top-side pressure (Prt)
We failed to control the top-side pressure directly using a PI controller.Different tunings were used without any success. However, we could stabilizethe system when we used the top-side pressure measurement to estimate thestates using a fast nonlinear observer, and applied a linear state feedback.We make two hypotheses to explain this observation:
� The top-side pressure dynamics is highly nonlinear towards the valve
� The RHP-zero (inverse response) in top pressure dynamics limits thecontrollability
One or both of these two reasons can be correct. We tried to stabilize thesystem by using a fast linear observer, but it was not possible. This con-firms that the first factor (nonlinearity effect) is important. The nonlinearobserver counteracts the nonlinearity of the dynamics and the states aremore linear towards the valve.
The unstable zero dynamics combined with unstable poles lead to highpeaks of the sensitivity transfer functions and they impose a limitation onthe controllability of the system. The minimum achievable peaks of sen-sitivity and complementary sensitivity transfer functions, denoted MS,min
and MT,min, respectively, are closely related to the distance between the un-stable poles (pi) and zeros (zi). Considering SISO systems, for any unstable(RHP) zero z:
‖S‖∞ ≥ MS,min =
Np∏
i=1
|z + pi||z − pi|
. (6.16)
This bound for our system at the operating point Z = 20% with the top-side pressure as the output is 2.1, which is quite high signals fundamentalproblem related to stabilization. However, this bound does not considerhow much fast control (bandwidth) is needed to stabilize the system. We
Table 6.3: State estimation using different observers
method \ measurement subsea pressure top pressure
fast linear observer working not workingfast nonlinear observer not working workingslow nonlinear observer working working
6.2. State Feedback with Nonlinear Observer 127
investigated this more closely using a H∞ controller for control of the top-side pressure. We need a fast control action, with about 1 sec responsetime, to stabilize the system. When applying a bandwidth of 1 rad/sec inloop shaping specifications of the H∞ controller, the peak of the sensitivitytransfer function increases to 4. This is called the water-bed effect ; if we pushdown the sensitivity transfer function in low frequencies, it will increase inother frequencies [74]. Furthermore, the minimum achievable peak of thesensitivity function increases to 7.0 for Z = 40%. This means that thestabilization with larger valve openings becomes impossible.
Measuring subsea pressure (Pin)
“Slow” nonlinear observers like the standard UKF can produce estimatesof the top-side pressure and its related states when measuring the subseapressure. However, when increasing the observer gain they fail. In contrary,the fast nonlinear observers work well when measuring the top-side pressure.what is the reason?
The structure of the model for the high-gain observer is in [51] introducedas a chain of integrators with the measured state is at the end of the chain(e.g. measuring the position and estimating velocity in mechanical systems).The high-gain observer behaves like a differentiator and the estimate ismore accurate with a higher gain, similar to calculating derivative by finitedifference by using smaller step [51].
If we consider the structure of the four-state model [34] with the topsidepressure as a measurement, we indeed find a chain of integrators. Themass of gas in the riser, which is closely related to the top-pressure, isthe integral of the gas flow rate at the riser-base, which is determined bythe subsea pressure. Roughly speaking, the subsea pressure is related toderivative of the top-side pressure, and as a result, fast observer works verywell by measuring the top pressure.
Table 6.4: Stabilizing control using different methods
method \ measurement subsea pressure top pressure
linear controller (PI, H∞) working not workingfast linear observer working not working
fast nonlinear observer not working workingslow nonlinear observer not robust not robust
max. valve opening 40% 20%
128 NONLINEAR CONTROL SOLUTIONS
On the other hand, by measuring the the subsea pressure, we have theopposite situations and we actually need to integrate the measured pressureto find the top-side pressure and the associated state variables. There aretwo problems arising in estimation by integration. First, for a good esti-mate we need a larger integration time; this is the reason for success of slowobservers in estimation. The second problem is that it is very sensitive tomodeling errors and disturbances. When an unknown model change or dis-turbance happens, the observer continues to integrate the wrong conditions,and finally it will fail as shown in Figure 6.8. In fact, even when we initializethe system in closed-loop to keep it always near the stationary point, it wasnot robust against disturbances in inflow conditions.
In the case of the linear observer, we assume the correct stationary pointand we only deal with the deviations. Consequently, as shown in Figure 6.9,the estimation of the linear observer is less accurate, but it does not fail tostabilize the system.
6.3 Output Linearizing Controller
The separation principle does not hold for nonlinear systems in general,and the closed-loop observer/controller is not guaranteed to be stable forall conditions. In the previous Section, we showed that a nonlinear observerfails in closed-loop for the supposedly simple case where the subsea pressureis the measurement. On the other hand, the system could be stabilized onlyin a limited range (Z = 20%) using a fast nonlinear observer and measuringthe topside pressure. Here, we propose a nonlinear model-based controllerby feedback linearization [38]. This controller directly uses the pressureat the riser-base (Prb) and the pressure at the top of the riser (Prt) asmeasurements, without using any observer.
6.3.1 Cascade system structure
In order to simplify the system analysis, we separate it into two subsystemsand analyze the individual subsystems and their interconnecting relation-ships. As illustrated in Fig. 6.10, the input to the “Riser” subsystem (Σ2)is the choke valve opening, u = Z, and the output is the pressure at theriser-base, Prb, which is also the input to the “Pipeline” subsystem (Σ1).
6.3.2 Stability analysis of cascade system
Apart from the riser-slugging, other phenomena may lead to the flow in-stability in pipeline-riser systems. The pipeline-riser system may have an
6.3. Output Linearizing Controller 129
Prb u
w
PipelineRiser
Figure 6.10: Pipeline-riser system as a cascaded connection of two sub-systems
unstable oil well as the inlet boundary, also the density-wave instability canhappen in long risers. Here, we consider only the riser-slugging instabilityand we state the following hypothesis about the “Pipeline” subsystem:
Hypothesis 1. If riser-slugging is the destabilizing dynamics of the pipeline-riser system, then the “Pipeline” subsystem with the riser-base pressure, Prb,as its input is“input-to-state stable”.
We investigate the input-to-state stability of the pipeline subsystem bya simulation test as shown in Fig. 6.11. The riser-base pressure in this sim-ulation is 19.7 kPa. This pressure is corresponding to 50% opening of thetop-side valve for which the pipeline-riser system is unstable. However, thepipeline subsystem separated from the riser is always stable. In addition,the local exponential stability of the Pipeline” subsystem can be verifiedby looking at eigenvalues. The above hypothesis is reasonably correct, be-cause the riser-base pressure is a recommended candidate controlled variableto stabilize the system [40]. This means when the riser-base pressure hassmall variations, the whole system becomes stabilized which follows L2-gainstability from Prb to state variables of the pipeline subsystem.
We can now state the following proposition:Proposition 1. Let hypothesis 1 hold. If the Riser subsystem becomes globallyasymptotically stable under a stabilizing feedback control, then the pipeline-riser system is globally asymptotically stable.
Proof : We use conditions for stability of cascaded systems as stated byCorollary 10.5.3 in [33]. As shown in Fig. 6.12, if Σ1 is input-to-state stable(ISS) and origin of Σ2 is globally asymptotically stable (GAS), then originof the cascaded system Σ1 and Σ2 is globally asymptotically stable (GAS).Therefore, if Hypothesis 1 holds, the proposition is verified.
�
It is not possible to achieve GAS for the riser subsystem due to controlla-bility limitations and other physical limits of the system. Instead, we showpartial stabilization with respect to the output that enters the pipeline sub-system, Prb, on a set D which is the physical domain of the system.
130 NONLINEAR CONTROL SOLUTIONS
0 50 100 150 200 250 3000.031
0.032
0.033
0.034
a) mass of gas in pipeline, mgp
mg
p [kg]
time [sec]
0 50 100 150 200 250 3000.82
0.825
0.83
b) mass of liquid in pipeline, mlp
mlp
[kg]
time [sec]
0 50 100 150 200 250 30018
19
20
21
c) pressure at riser−base as input, Prb
Prb
[kP
a ga
uge]
time [sec]
Figure 6.11: Simulation test of pipeline subsystem
P
GAS
w
1: ISS 2: GAS
Figure 6.12: Stability of a cascaded system
6.3. Output Linearizing Controller 131
6.3.3 Stabilizing Feedback Control
We use feedback linearization to design a control law. For simplicity, theoutlet mass flow rate, w, is used as a virtual control input. First, we trans-form the state equations and write a model in ‘normal form’ based on thetwo measurements, the pressure at the riser-base (Prb) and the pressure attop of the riser (Prt).
Transforming state equations
Two state variables govern the dynamics of the Riser subsystem (Σ2). Westart with writing the model in the new coordinates (x3, x4) = (mgr,mlr +mgr). We get the following system:
x3 = wg − αw (6.17)
x4 = wg +wl − w (6.18)
The two measurements are y1 = Prb and y2 = Prt. From the ideal gas lawwe get
y2 =ax3
b+ x3 − x4, (6.19)
where a = RTrρl/MG and b = ρlVr are model parameters [34]. The pressuredrop over the riser is the sum of the hydrostatic head and the friction term:
y1 = y2 + cx4 + Fr, (6.20)
where c = gLr/Vr and Fr is the friction in the riser that depends on constantinflow rates and other constant parameters. Differentiating and rearrangingthe equations gives the system equations in y coordinates.
y1 = (wg + wl) [F (y) + c]− [F (y) + c]w (6.21)
y2 = (wg + wl)F (y)− F (y)w, (6.22)
where
F (y) = c(
1− y2a
) aα+ y2(1− α)
bc− (y1 − y2 − Fr). (6.23)
Since ρl > ρg, we have that a > y2. In addition, bc = ρlgLr is the hydrostaticpressure when the riser is full of liquid which is larger than the gravity termin normal operation (y1−y2−Fr). Thus, the physical domain of the systemoutputs is defined as follows:
D = {(y1, y2)|y2 + bc+ Fr > y1 > y2 > 0}. (6.24)
132 NONLINEAR CONTROL SOLUTIONS
The numerator in equation (6.23) is always positive, therefore, F (y) >0,∀(y1, y2) ∈ D. In addition, it can be shown that F (y) is strictly increasingin y1 and strictly decreasing in y2.
The transformation T : S → D, y = T (x) in (6.19) and (6.20) wherey = (y1, y2)
T , x = (x3, x4)T and
T (x) =
[ ax3
b+x3−x4+ cx4 + Frax3
b+x3−x4
]
(6.25)
is a diffeomorphism on S = {(x3, x4)|x3 > 0, x4−x3 < b}, because both T (x)and T−1(y) exist and are continuously differentiable. b = ρlVr is the massof a volume of liquid equal to volume of the riser, hence b > mlr = x4 − x3in the normal operation of the system where x3 > 0.
For simplicity we will use the following assumption:Assumption 1. The gas mass fraction in the riser, α, is given by the constantinflow rates of the gas and the liquid:
α =wg,in
wg,in +wl,in. (6.26)
This particularly holds exactly at steady state, but dynamically it is a sim-plification.
Partial input-output linearization
With ξ = y1 − y1 (riser-base pressure) and η = y2 − y2 (topside pressure),where y1 and y2 are steady-state values, we can write the system in normalform:
ξ = (wg + wl) [F (ξ, η) + c]− [F (ξ, η) + c]w (6.27)
η = (wg + wl)F (ξ, η) − F (ξ, η)w, (6.28)
The feedback controller
w =−1
F (ξ, η) + c(−(wg +wl) [F (ξ, η) + c] + v) (6.29)
reduces equation (6.27) to ξ = v and choosing v = −K1ξ gives
ξ = −K1ξ, (6.30)
whereK1 > 0 results in exponential stability of the ξ dynamics. By insertingthe control law (6.29) into (6.28) we get
η =F (ξ, η)
F (ξ, η) + cv = −FK1ξ, (6.31)
6.3. Output Linearizing Controller 133
where
0 < F < F =F (ξ, η)
F (ξ, η) + c< F < 1. (6.32)
Since 0 < F < 1 and ξ → 0 known as exponentially fast, η will thereforeremain bounded. This is partial exponential stabilization of the system withrespects to the riser-base pressure ξ [81].
�
Assumption 2. In order to make the control law realizable, we replace wg+wl
by inflow rates to the system, wg,in + wl,in = win, such that
w =win(F (y) + c) +K1(y1 − y1)
F (y) + c(6.33)
The final control signal to the valve is
u = sat
(
w
Cv
√
ρrt(y2 − Ps)
)
, (6.34)
where Cv and Ps are the choke valve constant and the separator pressure,respectively. The riser friction Fr and the density ρrt are calculated basedon the two pressure measurements y1 and y2 and model parameters (seeAppendix D).
We can also design a control law that linearizes equation (6.28). Al-though we are using both y1 and y2 for F (y) in (6.28), we use only y2,the topside pressure, for feedback in the linear part of the controller. Theresulting controller is
w =1
F (y)(winF (y) +K2(y2 − y2)) , (6.35)
and the η-dynamics are exponentially stable. The final control signal tothe valve is same as equation (6.34). However, for feedback linearizationwe need the system to be minimum phase, but the linearized 4-state modelconstitutes two Right-Half-Plane zeros from the valve position (input) tothe top-side pressure (output) [40]. Although we cannot prove stability ofthe system using the controller in (6.35), we will try it in experiments.
Stability of composite system
After designing a stabilizing feedback control for the “Riser” subsystemusing feedback linearization, we consider the complete pipeline-riser systemas a partially linear composite system,
x = f(x, ξ), x ∈ R2, ξ ∈ D (6.36)
134 NONLINEAR CONTROL SOLUTIONS
ξ = −K1ξ, (6.37)
where f(x, ξ) represents dynamics of the “Pipeline” subsystem. We cancheck the conditions for stabilization of the composite system as stated by[64] and [42]:“A linear controllable nonlinear asymptotically stable cascade system is glob-ally stabilizable by smooth dynamic state feedback if (a) the linear subsystemis right invertible and weakly minimum phase, (b) the only variables enteringthe nonlinear subsystem are the outputs and the zero dynamics correspond-ing to this output.”
We use Hypothesis 1 for stability of the nonlinear (pipeline) subsystem.The two conditions (a) and (b) are satisfied when using the pressure atthe riser-base as the output. The riser-base pressure is minimum-phaseand it is the output which enters the nonlinear subsystem as shown inFig. 6.10. Therefore, the composite system is stabilizable on the domainD by using the riser-base pressure as the controlled output. However, thetop-side pressure is not minimum-phase and this output does not enter thenonlinear subsystem.
6.3.4 Analogy with conventional cascade control
The proposed control law is very similar to a conventional cascade con-troller that uses a flow controller as its inner loop and a pressure controlleras the outer loop [74]. The control law in equation (6.33) is a nonlinearpressure controller that specifies the set-point of the flow rate. In addition,equation (6.34) determines the valve opening based on the flow rate. Flowcontrollers are also used in conventional cascaded controllers to remove non-linearities of the valve. However, the difference is that we do not use anyflow measurement, instead the model estimates the flow rate from the twopressure measurements. The virtual control signal w can be considered asan estimate of the flow rate.
6.3.5 Experimental Results
The two pressure measurements at the riser-base and at top of the riser arevery noisy because of air bubbles and hydrodynamic slugs which have muchfaster dynamics than the severe-slugging dynamics. In order to reduce thenoise effect on the control signal (input), we used a second-order low-passfilter. The experimental result using the riser-base pressure for feedback-linearization with a valve opening Z = 30% is shown in Fig. 6.13. Thelow-pass filter was used for this case and the system could be stabilizedwithout much noise effect.
6.3. Output Linearizing Controller 135
0 2 4 6 8 10 12 14 16 18 2010
15
20
25
30
35
40
t [min]
Prb
[kP
a]
riser−base pressure (controlled variable)
open−loop stable
open−loop unstable
set−pointmeasurement
0 2 4 6 8 10 12 14 16 18 200
5
10
15 open−loop stable
open−loop unstable
topside pressure (measurement used by controller)
t [min]
Prt [k
Pa]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
100
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.13: Experimental result of using riser-base pressure (Prp) forfeedback linearization, Z = 30%
136 NONLINEAR CONTROL SOLUTIONS
It is desirable to open the top-side choke valve as much as possible toget the maximum production. For example, one experiment was performedwith a valve opening Z = 60% as shown Fig. 6.14. We did not need tore-tune the controller; the control law generates a larger proportional gainto stabilize the system. On the other hand, the low-pass filter adds a time-lag to the control loop, and it limits the controllability. The extra time-lagfrom the low-pass filter destabilises the closed-loop system when using thelarge control gain. Therefore, we can not filter out the noise completely; weshould keep the system stable in expense of accepting a noisy input signal.
In order to avoid the noise effect, we could use the buffer tank pressureinstead of the riser-base pressure. There are no bubbles in the buffer tankand its pressure is very close to the riser-base pressure. Indeed, althoughthe controller law was designed for the riser-base pressure, it works verywell using the buffer-tank pressure and with less noise effect as shown inFig. 6.15.
We could stabilize the system using the controller in (6.35) which hasthe top-side pressure in the linear part. The maximum achievable valveopening for this case as illustrated in Fig. 6.16 is 20%. This result is sameas using a nonlinear observer and state feedback [37].
For comparison, we carried out some experiment using PI controller.First, we considered using pressure drop of the riser (y1 − y2) which is themost simple combination of the two pressure measurements we used forfeedback-linearizing controller. The pressure drop over the riser was alsorecommended by [15] as the best control solution. As shown in Fig. 6.17,although the pressure drop over the riser is tightly controlled, both the riser-base pressure and the top-side pressure drift away in the same direction.Next, we used the riser-base pressure as the controlled variable of PI
controller for two valve openings of 30% and 50%. The results are given inFig. 6.18 and Fig. 6.19 respectively. However, we needed to re-tune the PIcontroller for the second valve opening, and Kc was changed from -10 to-20.
6.3.6 Controllability Limitations
When using the riser-base pressure for feedback linearization the controllercounteracts the nonlinearity of the system and we are able to stabilize thesystem for very large valve openings. The only limitation regarding theriser-base pressure is that with large valve openings the gain of the systemdecreases drastically and the controller generates a very large proportionalgain to stabilize the system. In this situation, the controller can not differ-entiate between noises and the unstable dynamics. The controller amplified
6.3. Output Linearizing Controller 137
0 2 4 6 8 10 12 14 16 18 2010
15
20
25
30
35
40
t [min]
Prb
[kP
a]
riser−base pressure (controlled variable)
open−loop stable
open−loop unstable
set−pointmeasurement
0 2 4 6 8 10 12 14 16 18 200
5
10
15 open−loop stable
open−loop unstable
topside pressure (measurement used by controller)
t [min]
Prt [k
Pa]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
100
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.14: Experimental result of using riser-base pressure (Prb) forfeedback linearization, Z = 60%
138 NONLINEAR CONTROL SOLUTIONS
0 2 4 6 8 10 12 14 16 18 2010
15
20
25
30
35
40
t [min]
Pin
[kP
a]
inlet pressure (controlled variable)
open−loop stable
open−loop unstable
set−pointmeasurement
0 2 4 6 8 10 12 14 16 18 200
5
10
15 open−loop stable
open−loop unstable
topside pressure (measurement used by controller)
t [min]
Prt [k
Pa]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
100
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.15: Experimental result of using buffer tank pressure (Pin) forfeedback linearization, Z = 60%
6.3. Output Linearizing Controller 139
0 2 4 6 8 10 12 14 16 18 2015
20
25
30
35
40
open−loop stable
open−loop unstable
riser−base pressure (measurement used by controller)
t [min]
Prb
[kP
a]
0 2 4 6 8 10 12 14 16 18 200
5
10
15
t [min]
Prt [k
Pa]
topside pressure (controlled variable)
open−loop stable
open−loop unstable
set−pointmeasurement
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
100
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.16: Experimental result of using top-side pressure (Prt) for feed-back linearization, Z = 20%
140 NONLINEAR CONTROL SOLUTIONS
0 2 4 6 8 10 12 14 16 18 2010
15
20
25
30
35
t [min]
Prb
−P
rt [k
Pa]
pressure drop over riser (controlled variable)
set−pointmeasurement
0 2 4 6 8 10 12 14 16 18 2010
20
30
40
open−loop stable
open−loop unstable
riser−base pressure (measurement used by controller)
t [min]
Prb
[kP
a]
0 2 4 6 8 10 12 14 16 18 200
5
10
15 open−loop stable
open−loop unstable
topside pressure (measurement used by controller)
t [min]
Prt [k
Pa]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
100
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.17: Experimental result of using pressure drop over riser (Prb −Prt) as controlled variable of a PI controller with Kc = −5,Ti = 120 s, Z = 40%
6.3. Output Linearizing Controller 141
0 2 4 6 8 10 12 14 16 18 2010
15
20
25
30
35
40
t [min]
Prb
[kP
a]
riser−base pressure (controlled variable)
open−loop stable
open−loop unstable
set−pointmeasurement
0 2 4 6 8 10 12 14 16 18 200
5
10
15 open−loop stable
open−loop unstable
topside pressure
t [min]
Prt [k
Pa]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
100
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.18: Experimental result of using riser base pressure (Prb) as con-trolled variable of a PI controller withKc = −10, Ti = 120 s,Z = 30%
142 NONLINEAR CONTROL SOLUTIONS
0 2 4 6 8 10 12 14 16 18 2010
15
20
25
30
35
40
t [min]
Prb
[kP
a]
riser−base pressure (controlled variable)
open−loop stable
open−loop unstable
set−pointmeasurement
0 2 4 6 8 10 12 14 16 18 200
5
10
15 open−loop stable
open−loop unstable
topside pressure
t [min]
Prt [k
Pa]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
100
Controller Off Controller On Controller Off
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.19: Experimental result of using riser base pressure (Prb) as con-trolled variable of a PI controller withKc = −20, Ti = 120 s,Z = 50%
6.4. PI Tuning Considering Nonlinearity 143
the noises and the control signal is very aggressive as shown in Fig. 6.14.One problem is how fast our valve can follow the control command signal,another problem is saturation of the valve.
By using the top-side pressure, linear controllers (PID, LQG, Hinf) arenot able to stabilize the system, but the nonlinear control law based on feed-back linearization could stabilize the system. However, we could stabilizethe system using the top-pressure in a very limited range (20% valve open-ing). Although the nonlinear controller law can counteract the nonlinearity,the fundamental limitations regarding the non-minimum phase dynamics[74] are still in place.
6.3.7 Remarks on output-linearizing controller
A nonlinear model-based output-linearizing controller was proposed for anti-slug control. The proof of convergence was shown in theory and experi-ments. The controller was able to stabilize the system up to very largevalve openings without re-tuning.
The advantage of the proposed controller over the previous works is thatit directly uses two pressure measurements at the riser-base and at the risertop, not the state variable. We do not have to deal with observers and hopefor the the separation principle to be applicable.
Furthermore, we showed that there are controllability limitations forthe system when using the riser-base pressure and the top pressure forthe feedback linearization design. The fundamental limitation related tousing the riser-base pressure is the small gain of the system with large valveopenings. In addition to nonlinearity, the top-side pressure has still thelimitation regarding non-minimum phase dynamics which can not be by-passed by any control solution.
6.4 PI Tuning Considering Nonlinearity
Here, we consider a PI controller and change the PI tuning values at differentoperating points to counteract the linearity of the system. This can beimplemented as a gain-scheduling controller or an adaptive controller. Thegain-scheduling or adaptation is based on the static gain of the system andthe average valve opening.
6.4.1 Simple model for static nonlinearity
The middle line in the bifurcations diagrams (Figure 6.3) represents thedesired non-slug flow. The slope of this line ( ∂y∂u = ∂P
∂Z ) represents the
144 NONLINEAR CONTROL SOLUTIONS
process gain. We found that this curve is mostly related to valve properties.We assume the valve equation as the following:
w = Cvf(z)√
ρ∆P (6.38)
where w[kg/s] is the outlet mass flow and ∆P [N/m2] is the pressure drop.From the valve equation, the pressure drop over the valve for different valveopenings can be written as
∆P =a
f(z)2, (6.39)
where we assume a as a constant parameter calculated in Appendix C. Oursimple empirical model for the inlet pressure is as follows:
Pin =a
f(z)2+ Pfo (6.40)
Where Pfo is another constant parameter that is the inlet pressure whenthe valve is fully open, and it is given in Appendix C. By differentiating(6.40) with respect to z, we get the static gain of the system as a functionof valve opening.
k(z) =−2a∂f(z)
z
f(z)3(6.41)
For a linear valve (i.e. f(z) = z) it reduces to
k(z) =−2a
z3, (6.42)
where 0 ≤ z ≤ 1. Figure 6.20 compares the simple static model in (6.40)and (6.41) to the Olga model.
6.4.2 PI tuning rules based on static model
The PID and PI tuning rules given in Chapter 5 are based on a linear modelidentified at a certain operating point. However, as we see in Figure 6.20,the gain of the system changes drastically with the valve opening. Hence, acontroller working at one operating point may not work at other operatingpoints.
One solution is gain-scheduling with multiple linear controllers based onmultiple identified models that we will use in the next Section. Here, wepropose simple PI tuning rules based on a single step test. We apply a gaincorrection to counteract the nonlinearity of the system. For this, we use the
6.4. PI Tuning Considering Nonlinearity 145
0 10 20 30 40 50 60 70 80 90 10065
70
75
80
Z [%]
Pin
[b
ar]
Olga model simple static model
0 10 20 30 40 50 60 70 80 90 100−4
−3
−2
−1
0
1
Z [%]
k(z)
Olga model simple static model
Figure 6.20: Simple static model compared to OLGA case
static model given in (6.41) and the data from a closed-loop step test as inFigure 6.21. We calculate the following constant:
β =− ln
(
∆y∞−∆yu∆yp−∆y∞
)
2∆t+
Kc0k(z0)(
∆yp−∆y∞∆y∞
)2
4tp, (6.43)
where z0 is the average valve opening in the closed-loop step test and Kc0
is the proportional gain used for the test. The PI tuning values as functionsof valve opening are given as the following:
Kc(z) =βTosc
k(z)√
z/z∗(6.44)
τI(z) = 3Tosc(z/z∗) (6.45)
Where Tosc is the period of slugging oscillations when the system is open-loop and z∗ is the critical valve opening of the system (at the bifurcationpoint).
6.4.3 Experimental result
First, we performed a closed-loop step test at Z = 20%(z0 = 0.2) where thepressure set-point is 26 kPa. We closed the loop by a proportional controllerKc0 = −10 and changed the set-point by 2 kPa (Figure 6.21). We calculated
146 NONLINEAR CONTROL SOLUTIONS
20 40 60 80 100 120 14025.5
26
26.5
27
27.5
28
28.5
29
time [sec]
y [kpa]
output
set point
!yp
!y!y
u
8
tpt
!ys
u
Figure 6.21: Closed-loop step response for stabilized experimental system
β = 0.061 from (6.43) using the step test information taken from Figure 6.21.The period of the slugging oscillations in open-loop was Tosc = 68 sec. Weused the PI tuning given in (6.44) and (6.45) in an adaptive manner tocontrol the system. The valve opening had many variations and it couldbe used directly; a low-pass filter was used to make it smooth. The resultof control using this tuning is shown in Figure 6.22. The controller gainsare given in Figure 6.23. This simple adaptive controller could stabilize thesystem from Z = 20% to Z = 50%, and it was stable even with 1sec addedtime delay. It can stabilize the system up to Z = 60% when no time delayis added.
6.4.4 Olga Simulation
Here, we test the PI tuning rules in (6.44) and (6.45) as a gain-schedulingon the Olga case presented in Section 2.3.1. The PI tuning values are givenin Table 6.5 and the simulation result is shown in Figure 6.24. The open-loop system switches to slugging flow at 5% valve opening, but by usingthe proposed PI tuning, the system could be stabilized up to 23.24% valveopening.
6.4. PI Tuning Considering Nonlinearity 147
0 5 10 15 20 25 3015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
Pin
[kp
a]
t [min]
0 5 10 15 20 25 300
20
40
60
80
Controller OffController On
Controller Off
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 6.22: Result of control using adaptive PI tuning in experiment 3
0 5 10 15 20 25 30−100
−80
−60
−40
−20
0open−loop stable
Kc [
−]
t [min]
proportional gain
0 5 10 15 20 25 30100
200
300
400
500
600
700
τ I [sec
]
t [min]
integral time
Figure 6.23: Controller gains resulted from adaptive PI tuning in exper-iment 3
148 NONLINEAR CONTROL SOLUTIONS
0 1 2 3 4 5 666.5
67
67.5
t [h]
Pin
[b
ar]
inlet pressure (controlled variable)
measurement set−point
0 1 2 3 4 5 60
10
20
30
40
50
Z [
%]
t [h]
valve position (manipulated variable)
Figure 6.24: Result of control from Olga simulation
6.5 Gain-Scheduling IMC
In Section 6.3, we applied feedback linearization for anti-slug control. How-ever, the feedback linearization design is not robust against modeling errors,since the nonlinearities of the system must be perfectly known in order tocancel them.
An alternative approach in which the mechanistic model is not directlyused for the control design is to identify an unstable model of the systemby a closed-loop step test. We use the subsea pressure measurement as the
Table 6.5: PI tuning values in Olga simulation
set-point valve opening Kc τI67.36 14 0.5 840067.19 16.1 0.7 960067.07 18.2 0.94 1080066.99 20.1 1.23 1200066.93 23.24 1.56 1320066.88 – 1.93 14400
6.5. Gain-Scheduling IMC 149
controlled output, and the identified model contains two unstable poles andone stable zero.
G(s) =b1s+ b0
s2 − a1s+ a0=
k′(s+ ϕ)
(s− π1)(s − π2)(6.46)
We use the identified model for an IMC (Internal Model Control) design[58]. The feedback version of the IMC controller is in the following form.
C(s) =[ 1k′λ3 ](α2s
2 + α1s+ 1)
s(s+ ϕ). (6.47)
Where λ is an adjustable filter time-constant. The filter coefficients α1 andα2 are calculated by solving the following system of linear equations:
(
π12 π1 1
π22 π2 1
)
α2
α1
α0
=
(
(λπ1 + 1)3
(λπ2 + 1)3
)
(6.48)
Where we choose α0 = 1 to get an integral action in the controller. Theclosed-loop model identification and the IMC design for the slugging flowsystem have been presented in Chapter 5 in more details.
Here, we identify three linear models from step tests at three differentoperating points, Z = 20%, Z = 30% and Z = 40%, respectively,
G1(s) =−0.015(s + 0.26)
s2 − 0.045s + 0.0094, (6.49)
G2(s) =−0.0098(s + 0.25)
s2 − 0.040s + 0.025, (6.50)
G3(s) =−0.0056(s + 0.27)
s2 − 0.017s + 0.096. (6.51)
Then, we design three IMC controllers to cover operation range of the non-linear system for small, medium and large valve openings.
C1(s) =−16.15(s2 + 0.016s + 0.0012)
s(s + 0.26)(6.52)
C2(s) =−42.20(s2 + 0.052s + 0.0047)
s(s + 0.25)(6.53)
C3(s) =−115.11(s2 + 0.052s + 0.014)
s(s + 0.27)(6.54)
Switching (gain-scheduling) between the three controllers is based on thepressure set-point as given in Table 6.6, and bump-less transfers betweencontrollers are considered.
150 NONLINEAR CONTROL SOLUTIONS
Table 6.6: Gain-scheduling logic
Pressure set-point Controller
Pset ≥ 24 kPa C1(s)24 kPa > Pset > 21.5 kPa C2(s)
Pset ≤ 21.5 kPa C3(s)
6.6 Comparing Four Nonlinear Controllers
We compare the four nonlinear controllers presented in the previous sectionsexperimentally. All the experiments are performed on set of descendingpressure set-points to observe where the system becomes unstable. A lowerpressure set-point which gives a larger valve opening is desirable, but itis more difficult to control. To test the robustness of the controllers, werepeated each experiments for three different values of time delay 1 sec, 2sec and 3 sec, but we do not show results for 1 sec time-day because ofspace limitation.
Figure 6.25 shows result of using the state-feedback/nonlinear observerscheme for control. It can stabilize the system up to 28% valve opening.However, with 2 sec time delay, as in Figure 6.26, it is stable only up to22% valve opening.
The observer/state-feedback scheme can stabilize the system only whenusing the top-side pressure Prt as the measurement for the observer, sincea high-gain observer diverges when using the subsea pressure Pin as themeasurement (see Figure 6.8). The reason for this has been explained inSection 6.2.4.
Figure 6.27 shows the experimental result of the adaptive PI controllerwith 2 sec time delay. This controller was able to stabilize the system upto Z = 60% when no time delay was added, and Z = 32% with 2 sec addedtime delay.
Figure 6.28 shows result of using the feedback linearization controller.With no time delay, it stabilizes the system up to 60% valve opening. How-ever, with 2 sec time delay, as in Figure 6.29, it is stable only up to 25%valve opening.
Figure 6.30 shows result of applying gain scheduling IMC. This schemestabilizes the system up to 60% valve opening, and even with 2 sec timedelay (Figure 6.31), it is stable up to 50% valve opening.
Table 6.7 shows the maximum valve opening achieved by using the fourcontrollers with different values of time-delay.
6.6. Comparing Four Nonlinear Controllers 151
0 5 10 15 20 25 3015
20
25
30
35
40
t [min]
Pin
[kP
a]
inlet pressure (estimated by observer)
open−loop stable
open−loop unstable
actualobserverset−point
0 5 10 15 20 25 300
5
10
15
t [min]
Prt [k
Pa]
topside pressure (measurement used by observer)
open−loop stable
open−loop unstable
actualobserver
0 5 10 15 20 25 300
20
40
60
80
100
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.25: Control using High-Gain observer measuring top pressure(Prt) with 0 sec time delay
152 NONLINEAR CONTROL SOLUTIONS
0 5 10 15 20 25 3015
20
25
30
35
40
t [min]
Pin
[kP
a]
inlet pressure (estimated by observer)
open−loop stable
open−loop unstable
actualobserverset−point
0 5 10 15 20 25 300
5
10
15
t [min]
Prt [k
Pa]
topside pressure (measurement used by observer)
open−loop stable
open−loop unstable
actualobserver
0 5 10 15 20 25 300
20
40
60
80
100
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.26: Control using High-Gain observer measuring top pressure(Prt) with 2 sec time delay
6.6. Comparing Four Nonlinear Controllers 153
0 2 4 6 8 10 12 14 16 18 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)P
in [
kpa]
t [min]
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
open−loop stable
open−loop unstable
Zm [
%]
t [min]
actual valve position (manipulated variable)
Figure 6.27: Control using adaptive PI controller measuring inlet pressure(Pin) with 2 sec time delay
Table 6.7: Maximum valve opening achieved by using different controllersand different values of time delay
CV θ = 0 θ = 1 θ = 2
Gain-scheduling IMC Pin 60% 60% 50%Adaptive PI Pin 60% 50% 32%
Output linearization Prb & Prt 60% 40% 25%Nonlinear observer Prt 28% 24% 22%
154 NONLINEAR CONTROL SOLUTIONS
0 5 10 15 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
t [min]
Pin
[kP
a]
0 5 10 15 200
5
10
15 open−loop stable
open−loop unstable
topside pressure
t [min]
Prt [k
Pa]
0 5 10 15 200
20
40
60
80
100
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.28: Control using nonlinear controller measuring subsea pressure(Pin) with 0 sec time delay
6.6. Comparing Four Nonlinear Controllers 155
0 5 10 15 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
t [min]
Pin
[kP
a]
0 5 10 15 200
5
10
15 open−loop stable
open−loop unstable
topside pressure
t [min]
Prt [k
Pa]
0 5 10 15 200
20
40
60
80
100
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.29: Control using nonlinear controller measuring subsea pressure(Pin) with 2sec time delay
156 NONLINEAR CONTROL SOLUTIONS
0 5 10 15 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
t [min]
Pin
[kP
a]
0 5 10 15 200
5
10
15 open−loop stable
open−loop unstable
topside pressure
t [min]
Prt [k
Pa]
0 5 10 15 200
20
40
60
80
100
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.30: Control using IMC controller measuring subsea pressure(Pin) with 0 sec time delay
6.6. Comparing Four Nonlinear Controllers 157
0 5 10 15 2015
20
25
30
35
40
open−loop stable
open−loop unstable
inlet pressure (controlled variable)
t [min]
Pin
[kP
a]
0 5 10 15 200
5
10
15 open−loop stable
open−loop unstable
topside pressure
t [min]
Prt [k
Pa]
0 5 10 15 200
20
40
60
80
100
open−loop stable
open−loop unstable
t [min]
Zm [%
]
actual valve position (manipulated variable)
Figure 6.31: Control using IMC controller measuring subsea pressure(Pin) with 2 sec time delay
158 NONLINEAR CONTROL SOLUTIONS
6.7 Summary
Four nonlinear anti-slug controllers were tested on the same experimentalplatform under the same conditions. First, we considered two nonlinearcontrollers based on the mechanistic model, the state-feedback/nonlinearobserver and the feedback linearizing controller. Then, we applied PI controlwith gain adaptation based on a simple model of the static gain. Finally, again-scheduling with IMC controllers based on identified models was used.The gain-scheduling IMC was able to stabilize the system up to large valveopenings (Z = 60%) even when applying time-delay to control loop (Z =50%); time-delay in the measurement is a major problem of long flow-lines.The IMC controller takes advantage of derivative action which results in abetter phase-margin for stabilizing control, while the two other nonlinearcontrollers (based on the mechanistic model) are essentially proportionalcontrollers. Other advantages of the IMC scheme are its simplicity and thefact that no mechanistic model is required for the controller design.
The second best controller for this case-study was the adaptive PI tuningthat stabilizes the system up to Z = 60% without delay and Z = 32% with2 sec delay. the third rank solution was output-linearizing controller whichuses two pressure measurements directly, without observer. This controllercould stabilize the system for large valve openings (Z = 60%), but it wasnot robust against time delay.
The high-gain observer was only applicable by using the top-side pres-sure measurement, in a limited range (Z = 28%). It was not stable whenusing the the subsea pressure measurement in closed loop.
Chapter 7
CONCLUSIONS AND
FUTURE WORK
7.1 Main Contributions
This thesis reports results of the research carried out to improve anti-slugcontrol at offshore oilfields. This research includes modeling, simulationsusing the OLGA simulator and experiments. The modeling and controlstructure design have been accomplished for three different offshore pro-cesses which severe-slugging flow occurs in (the pipeline-riser systems, thewell-pipeline-riser systems and the gas-lifted oil wells). We mainly focusedon the pipeline-riser systems in the control part.
The main contributions of the thesis are described below.
7.1.1 Modeling
In Chapter 2, a new four-state model was proposed for the severe-sluggingflow in the pipeline-riser systems. The model was compared to five othersimple models from the previous works, results from the OLGA simulators,as well as experiments. The proposed model matches well with the OLGAsimulations and the experiments. Then, we extended this model to the well-pipeline-riser systems. We compared also the extended model to results fromthe OLGA simulator that resulted in a good agreement. Furthermore, wepresented a modified three-state model for casing-heading instability in gas-lifted oil well and compared to the OLGA simulations. The gas-lift modelis not a major contribution of this thesis and we present it in Appendix A.
159
160 CONCLUSIONS AND FUTURE WORK
7.1.2 Control structure design
We focused on finding suitable controlled variables and manipulated vari-ables for anti-slug control in Chapter 3 and 4, respectively. We performedthe controllability analysis for the three main processes mentioned above.With the controllability analysis we found the fundamental limitations inanti-slug stabilizing control.
When using the top-side pressure for the anti-slug control, the RHP-plane zeros of the system with this output limit the controllability by largepeak on the sensitivity and the complementary sensitivity transfer functions.This problem becomes problematic for larger valve openings. As a result,using the top-side pressure, we can stabilize the system in a limited range.For example, in the experiments the system switches from non-slug flow toslug flow at Z∗
1 = 15%, and by using the topside pressure we could stabilizethe system only up to Z1 = 20%. Indeed, this limitation was independentfrom the controller that we use and even with nonlinear control solutionsthis limitation cannot be bypassed.
When using a pressure measurement from the seabed the system canbe stabilized in a large range (Z1 = 60% in experiments). However, thisrequires a large controller gain for large valve openings. We face two fun-damental limitations when controlling a subsea pressure with large valveopenings:
� Input saturation:The gain of the system (G) decreases and finally approaches to zero asthe valve opening increases. The loop gain (L = KG) should not varymuch for different operating points, therefore, the control gain (K)must be increased to stabilize the system with large valve openings.
u = KS(r −Gdd− n)
By using an integral action we are at the steady-state where the com-plementary sensitivity transfer function is T ≈ 1.
KS = G−1T ≈ G−1
This means that small G translates to a large KS and leads largeinputs due to measurement noises or disturbances.
� Time delay:Time delay causes problem especially when controlling with a largevalve opening, because the instability moves further away from the ori-gin (“faster instability”) for larger valve openings. One of poles moves
7.1. Main Contributions 161
further into the RHP as the valve opening increases and the term epθ
becomes larger and the peak on the complementary sensitivity trans-fer function increases. In addition, by increasing the controller gain(K) the delay margin of the control loop (θM ) decreases, because thecrossover frequency ωc increases for a large controller gain (fast controlaction).
θM = PM/ωc
�
In Chapter 4, we tested three different manipulated variable (different lo-cations for the control valve) for anti-slug control in a well-pipeline-risersystem. We found that a subsea valve located near the riser base is suit-able for anti-slug control and its operation range is same as the top-sidevalve. However, a subsea wellhead valve can not be used to prevent theriser slugging.
For the gas-lifted oil wells, the bottom-hole pressure is the best controlledvariable when using the production choke for control. However, the bottom-hole pressure is not measured directly. In addition, we found that combingtwo topside pressure (the annulus pressure and the tubing pressure) givessatisfactory result. Further, we investigated using the topside gas-lift valvefor the control purpose, but it was not an effective manipulated variable forstabilizing control.
7.1.3 PID and PI tuning
In Chapter 5, we proposed PID and PI tuning rules for robust anti-slugcontrol. We found from an order truncation on the four-state model thata second order model is enough for the control design. We considered amodel structure with one zero and two unstable poles, then we estimatedparameters of the unstable model from a closed-loop step test.
Then, we used the identified unstable model for an IMC design. Finally,the PID and the PI tuning parameters were obtained from the resulted IMCcontroller. The proposed tuning rules were tested in OLGA simulationsand experiments on two experimental rigs which showed applicability androbustness of the proposed tuning rules.
7.1.4 Nonlinear control
Three nonlinear control solutions were tested in Chapter 6. First, we usedstate-feedback with state estimation by nonlinear observers. Three types
162 CONCLUSIONS AND FUTURE WORK
of nonlinear observers were tested by experiments. Secondly, we appliedfeedback linearization with measured outputs. Finally, we designed a gain-scheduling IMC (Internal Model Control) based on linear models identifiedfrom closed-loop step test. We compared these three approaches in termsof robustness (delay margin of the control loop) and operation range (max-imum achievable valve opening) in experiments.
Both the gain-scheduling and the feedback-linearization solutions wereable to stabilize the system up to Z1 = 60%. However, the gain-schedulingIMC was the preferred solution. It does not need a physical model of thesystem and was more robust against time-delay in the subsea pressure mea-surement.
The state-feedback/solution was only applicable by using the top-sidepressure measurement where we have fundamental limitation of the RHP-zero dynamics. We could stabilize the system by this solution in a limitedrange of Z1 = 20%. For the stabilizing control we need a fast observer, butfast nonlinear observers diverge when using the the subsea pressure as themeasurement. Consequently, the closed-loop system with a fast nonlinearobserver and using subsea pressure measurement was not stable.
7.2 Future works
We have used a stable vertical oil well in the well-pipeline-riser case studyin order that we could focus on the riser-slugging instability. An extensionto this can be using an unstable oil well in the well-pipeline-riser case study.Oil wells can have different geometries other than simply vertical. Somegeometries may include low-points that causes flow instabilities in the oilwell. In addition, wells with low reservoir pressures and heavy oils are moreprone to flow instabilities.
Appendix A
GAS-LIFTED OIL WELL
MODEL
Here, we present a modified three-state model for casing-heading instabilityin gas-lifted oil wells. The model is very similar to the ones used in [1],[2], [18], [19], [20]. We have improved the model by considering a frictionterm for the pressure drop and the gas production from the oil well. Aschematic illustration of a gas-lifted oil wells is shown in Fig. A.1. Thestate variables are the mass of gas in the annulus (mg,a), the mass of gas inthe tubing (mg,t) and the mass of liquid in the tubing (ml,t). We consideralso production of gas from the reservoir, which gives the following massbalances.
dmg,a
dt= wg,in − wg,inj (A.1)
dmg,t
dt= wg,inj + wg,res − wg,out (A.2)
dml,t
dt= wl,res −wl,out (A.3)
In this model, wg,in is the mass flow rate of inlet gas to the annulus andwg,inj is the mass flow of injected gas from the annulus into the tubing.wg,res and wl,res are gas and liquid mass flow rates from the reservoir to thetubing. wg,out and wl,out are the mass flow rates of gas and oil outlet fromthe tubing, respectively.There is only gas phase inside the annulus, and pressure at top of the annuluscan be calculated by ideal gas law.
Pat =RTamg,a
MgVa(A.4)
163
164 GAS-LIFTED OIL WELL MODEL
Gas lift
choke
Production
choke
Oil outlet
Gas inlet
Annulus
Injection
valve
Tubing
Pres
Pgs
Ptt m,tt l,tt
Pat
Pbh
u1
u2wg,in
wout
Pab
Figure A.1: Schematic presentation of gas-lift oil well
165
Then, the pressure at bottom of the annulus is given by
Pab = Pat +mg,agLa
Va, (A.5)
Thus, the density of the gas phase at this point is
ρg,ab =PabMg
RTa. (A.6)
The inlet gas to the annulus comes from a source tank or a compressor withthe pressure Pgs, and the density of gas through the gas-lift choke can bewritten as:
ρg,in =PgsMg
RTa(A.7)
Therefore, gas mass flow into the annulus is
wg,in = Kgsu2
√
ρg,inmax(Pgs − Pat, 0). (A.8)
Because of high pressure, the fluid from the reservoir is saturated ([3]).Hence, we assume that distance between the bottom-hole and the injectionpoint, Lbh, is filled by liquid phase. This must be accounted for in calculatingthe volume of gas in the tubing. Consequently, the density of gas inside thetubing follows as
ρg,t =mg,t
Vt + SbhLbh −ml,t/ρl. (A.9)
Pressure at top of tubing using ideal gas law:
Ptt =ρg,tRTt
Mg(A.10)
Average mixture density inside tubing:
ρm =mg,t +ml,t − ρlSbhLbh
Vt(A.11)
Average liquid volume fraction inside tubing:
αl,t =ml,t − ρlSbhLbh
Vtρl(A.12)
gor is the constant mass ratio of gas and liquid produced from the reservoir,and gas mass fraction at bottom of the tubing is
αmg,b = gor/(gor + 1). (A.13)
166 GAS-LIFTED OIL WELL MODEL
Before calculating the inlet mass flow rate from the reservoir by use of thebottom-hole pressure in equation (A.27), the pressure drop due to friction isneeded to determine the bottom-hole pressure. However, we need to knowthe inlet flow rate to calculate the friction term. We evade this problemby using the nominal production rate of the well, wnom, in calculation offriction terms.Average superficial velocity of liquid phase in tubing:
Usl,t =4(1 − αm
g,b)wnom
ρlπD2t
(A.14)
Average superficial velocity of gas phase:
Usg,t =4(wg,in + αm
g,bwnom)
ρg,abπD2t
(A.15)
We have not calculated flow rate of the injected gas from the annulus intothe tubing yet, instead we use wg,in in equation (A.15); we believe averagesof these two variables are equal.Average mixture velocity in tubing:
Um,t = Usl,t + Usg,t (A.16)
Reynolds number of flow in tubing:
Ret =ρmUm,tDt
µ(A.17)
The Haaland equation [26] is used as the friction factor in the tubing.
1√λt
= −1.8 log10
[
(
ǫ/Dt
3.7
)1.11
+6.9
Ret
]
(A.18)
Pressure loss due to friction in tubing:
∆Pft =αl,tλtρmixU
2m,tLt
2Dt(A.19)
Pressure at bottom of tubing where gas being injected from annulus:
Ptb = Ptt + ρmgLt +∆Pft (A.20)
Mass flow rate of gas injected into tubing:
wg,inj = Kinj
√
ρg,abmax(Pab − Ptb, 0) (A.21)
167
Liquid velocity at bottom-hole:
Ul,b =wnom
ρlSbh(A.22)
Reynolds number of flow at bottom-hole:
Reb =ρlUl,bDb
µ(A.23)
Friction factor at bottom-hole:
1√λb
= −1.8 log10
[
(
ǫ/Db
3.7
)1.11
+6.9
Reb
]
(A.24)
Pressure loss due to friction from bottom-hole to injection point:
∆Pfb =λbρlU
2l,bLbh
2Db(A.25)
Pressure at bottom-hole:
Pbh = Ptb + ρlgLbh +∆Pfb (A.26)
Mass flow rate from reservoir to tubing:
wres = PImax(Pres − Pbh, 0) (A.27)
Mass flow rate of liquid from reservoir to tubing:
wl,res = (1− αmg,b)wres (A.28)
Mass flow rate of gas from reservoir to the well:
wg,res = αmg,bwres (A.29)
Density of gas at bottom of tubing:
ρg,tb =PtbMG
RTt(A.30)
Liquid volume fraction at bottom of tubing:
αl,tb =wl,resρg,tb
wl,resρg,tb + (wg,inj + wg,res)ρl(A.31)
168 GAS-LIFTED OIL WELL MODEL
With the same assumptions used for the phase fraction of the riser in Sec-tion 2.2.8, liquid volume fraction at top of the tubing can be written as
αl,tt = 2αl,t − αl,tb, (A.32)
Then, the mixture density at top of the tubing will be
ρtt = αl,ttρl + (1− αl,tt)ρg,t. (A.33)
Mass flow rate of mixture from production choke:
wout = Kpru1√
ρttmax(Ptt − P0, 0) (A.34)
Volumetric flow rate of production choke:
Qout = wout/ρtt (A.35)
Gas mass fraction at top of tubing:
αmg,tt =
(1− αl,tt)ρg,tαl,ttρl + (1− αl,tt)ρg,t
(A.36)
Mass flow rate of outlet gas from tubing:
wg,out = αmg,ttwout (A.37)
Mass flow rate of outlet liquid from tubing:
wl,out = (1− αmg,tt)wout (A.38)
�
The simplified model was fitted to a test case implemented in the OLGAsimulator. Constants and parameters used in the model are given in Ta-ble A.1. The stability map of the system is shown in Figure A.2 wherestability transitions of the OLGA model and the simplified model are com-pared. It was not possible to add the gas-lift choke to the OLGA model,therefore we used a constant gas source equal to wG,in = 0.8 [kg/s] in theOLGA model and we fitted the model with the constant gas rate. Then, weadded the gas-lift choke valve to the Matlab model so that the simplifiedmodel gives wG,in = 0.8 [kg/s] when the gas-lift choke opening is u2 = 0.4and the production choke opening is u1 = 0.3. The system switches fromstable to unstable at this operating point. Figure A.3 Shows comparison ofthe bifurcations diagrams of the system where the production valve openingu1 is considered as the degree of freedom while the gas-lift choke valve iskept constant at u2 = 0.4.
169
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
production choke opening (u1)
inje
cted
gas
rat
e (w
G,in
)
Stable
Unstable
model stability transitionOLGA stability transition
Figure A.2: Stability transition of system, blue markers for OLGA andred markers for simplified model
0 20 40 60 80 1000
50
100
150
200
a) Pressure at bottom−hole (Pbh
)
Pb
h [ba
r]
Prod. valve, u1 [%]
0 20 40 60 80 1000
20
40
60
80
100
b) Mass production flow rate (wout
)
wo
ut [k
g/s]
Prod. valve, u1 [%]
0 20 40 60 80 100
60
80
100
120
140
160
180
c) Pressure at bottom of annulus (Pab
)
Pa
b [bar
]
Prod. valve, u1 [%]
0 20 40 60 80 10050
100
150
d) Pressure at top of annulus (Pat
)
Pa
t [bar
]
Prod. valve, u1 [%]
0 20 40 60 80 100
20
40
60
80
e) Pressure at top of tubing (Ptt)
Ptt [b
ar]
Prod. valve, u1 [%]
0 20 40 60 80 100
100
200
300
400
500
600
f) Outlet volumetric flow rate (Qout
)
Qou
t [L/s
]
Prod. valve, u1 [%]
Figure A.3: Bifurcation diagrams of simplified gas-lift model (solid lines)compared to OLGA model (dashed lines)
170 GAS-LIFTED OIL WELL MODEL
Table A.1: Parameter values for gas-lift model
Symb. Description Values Units
R universal gas constant 8314 J/(kmol.K)g gravity 9.81 m/s2
µ viscosity 3.64× 10−3 Pa.sρl liquid density 760 kg/m3
Mg gas molecular weight 16.7 grTa annulus temperature 348 KVa annulus volume 64.34 m3
La annulus length 2048 m3
Pgs gas source pressure 140 barVt tubing volume 25.03 m3
Sbh cross-section below injection point 0.0314 m2
Lbh length below injection point 75 mTt tubing temperature 369.4 Kgor mass gas oil ratio 0 –Pres reservoir pressure 160 barwnom nominal mass flow from reservoir 18 kg/sDt tubing diameter 0.134 mLt tubing length 2048 mPI productivity index 2.47e-6 kg/(s.Pa)Kgs gas-lift choke cons. 9.98× 10−5 –Kinj injection valve cons. 1.40× 10−4 –Kpr production choke cons. 2.90× 10−3 –
Appendix B
MODEL IDENTIFICATION
CALCULATIONS
Stable closed-loop transfer function:
y(s)
ys(s)=
K2(1 + τzs)
τ2s2 + 2ζτs+ 1(B.1)
The Laplace inverse (time-domain) of the transfer function in (B.1) is givenin [83] as
y(t) = ∆ysK2 [1 +D exp(−ζt/τ) sin(Et+ φ)] , (B.2)
where
D =
[
1− 2ζτzτ +
(
τzτ
)2]
1
2
√
1− ζ2(B.3)
E =
√
1− ζ2
τ(B.4)
φ = tan−1
[
τ√
1− ζ2
ζτ − τz
]
(B.5)
By differentiating (B.2) with respect to time and setting the derivative equa-tion to zero, one gets time of the first peak:
tp =tan−1
(
1−ζ2
ζ
)
+ π − φ√
1− ζ2/τ(B.6)
And the time between the first peak (overshoot) and the undershoot:
tu = πτ/√
1− ζ2 (B.7)
171
172 MODEL IDENTIFICATION CALCULATIONS
The damping ratio ζ can be estimated as
ζ =− ln v
√
π2 + (ln v)2(B.8)
where
v =∆y∞ −∆yu∆yp −∆y∞
(B.9)
Then, using equation (B.7) we get
τ =tu
√
1− ζ2
π. (B.10)
The steady-state gain of the closed-loop system is estimated as
K2 =∆y∞∆ys
. (B.11)
We use time of the peak tp and (B.6) to get an estimate of φ :
φ = tan−1
[
1− ζ2
ζ
]
−tp
√
1− ζ2
τ(B.12)
From (B.4), we get
E =
√
1− ζ2
τ(B.13)
The overshoot is defined as
D0 =∆yp −∆y∞
∆y∞. (B.14)
By evaluating (B.2) at time of peak tp we get
∆yp = ∆ysK2
[
1 + D exp(−ζtp/τ ) sin(Etp + φ)]
(B.15)
Combining equation (B.11), (B.14) and (B.15) gives
D =D0
exp(−ζtp/τ ) sin(Etp + φ). (B.16)
We can estimate the last parameter by solving (B.3):
τz = ξτ +
√
ζ2τ2 − τ2[
1− D2(1− ζ2)]
(B.17)
173
Then, we back-calculate to parameters of the open-loop unstable model.The steady-state gain of the open-loop model is
K =∆y∞
Kc0 |∆ys −∆y∞| (B.18)
From this, we can estimate the four model parameters in equation (B.5) are
a0 =1
τ2(1 +Kc0Kp)(B.19)
b0 = Kpa0 (B.20)
b1 =K2τzKc0τ2
(B.21)
a1 = −2ζ/τ +Kc0b1, (B.22)
where a1 > 0 gives an unstable system.
Appendix C
STATIC NONLINEARITY
PARAMETERS
From equation (6.38) we have
a =1
ρ
(
w
Cv
)2
(C.1)
Where Cv is the known valve constant, w is the steady-state average outletflow rate and ρ is the steady-state average mixture density. The averageoutlet mass flow is approximated by constant inflow rates.
w = wg,in + wl,in (C.2)
In order to estimate the average mixture density ρ, we perform the followingcalculations, assuming a fully open valve.Average gas mass fraction:
α =wg,in
wg,in + wl,in(C.3)
Average gas density at top of the riser from ideal gas law:
ρg =(Ps +∆Pv,min)Mg
RT(C.4)
where Ps is the constant separator pressure, and ∆Pv,min is the (minimum)pressure drop across the valve that exists with a fully open valve. In thenumerical simulations ∆Pv,min is assumed to be zero but in our experimentsit was 2 kPa.Liquid volume fraction:
αl =(1− α)ρg
(1− α)ρg + αρL. (C.5)
175
176 STATIC NONLINEARITY PARAMETERS
Average mixture density:
ρ = αlρl + (1− αl)ρg (C.6)
In order to calculate the constant parameters Pfo in (6.40), we use the factthat if the inlet pressure is large enough to overcome a riser full of liquid,slugging will not happen. [78] used the same concept for stability analysis,also this was observed in our experiments. we define the critical pressure as
P ∗
in = ρLgLr + Ps +∆Pv,min (C.7)
This pressure is associated with the critical valve opening at the bifurcationpoint z∗. From (6.40), we get Pfo as the following:
Pfo = P ∗
in − a
f(z∗)2(C.8)
Appendix D
CALCULATION OF
FRICTION AND DENSITY
Here, we calculate the friction and the density which are need by the non-linear controller in equation (6.34).
Gas mass fraction:α =
wg,in
wg,in + wl,in(D.1)
Gas density:
ρg =y2MG
RTr(D.2)
Liquid volume fraction:
αlt =ρgwl
ρgwl + ρlwg(D.3)
Mixture density at top of riser:
ρrt = αlt.ρl + (1− αlt)ρg (D.4)
Liquid superficial velocity:
Usl =wl,in
ρlAr(D.5)
Gas superficial velocity:
Usg =wg,in
ρgAr(D.6)
Mixture velocity:Um = Usl + Usg (D.7)
Average density in riser:
ρ =y1 − y2cLrAr
(D.8)
177
178 CALCULATION OF FRICTION AND DENSITY
Reynolds number of flow in riser:
Re =ρUmDr
µ(D.9)
An explicit approximation of the implicit Colebrook-White equation pro-posed by [26] is used as the friction factor in the riser.
1√λ= −1.8 log10
[
(
ǫ/Dr
3.7
)1.11
+6.9
Re
]
(D.10)
Pressure loss due to friction in riser:
Fr =αltλρU
2mLr
2Dr(D.11)
REFERENCES
[1] Aamo, O., Eikrem, G., Siahaan, H., Foss, B., 2005. Observer design formultiphase flow in vertical pipes with gas-lift - theory and experiments.Journal of Process Control 15 (3), 247 – 257.
[2] Aamo, O. M., Eikrem, G. O., Siahaan, H., Foss, B., 2004. Observerdesign for gas lifted oil wells. In: American Control Conference. pp.1552–1557.
[3] Ahmed, T., 2006. Reservoir Engineering Handbook, Third Edition. El-sevier, Oxford, UK.
[4] Bai, Y., Bai, Q., 2005. Subsea Pipelines and Riser. Elsevier ScienceLtd, Oxford.
[5] Bendlksen, K. H., Malnes, D., Moe, R., Nuland, S., 1991. Dynamic two-fluid model olga. theory and application. SPE Production Engineering6 (2), 171–180.
[6] Boizot, N., Busvelle, E., Gauthier, J.-P., 2010. An adaptive high-gainobserver for nonlinear systems. Automatica 46 (9), 1483–1488.
[7] Brill, J. P., Beggs, H. D., 1991. Two-Phase Flow In Pipes, 6th Edition,Third Printing. University of Tulsa, Tulsa, Oklahoma.
[8] Chen, J., June 2000. Logarithmic integrals, interpolation bounds andperformance limitations in mimo feedback systems. IEEE Transactionson Automatic Control 45 (6), 1098–1115.
[9] Cosse, R., 1993. Basics Of Reservoir Engineering. Institut francais dupetrole (IFP), Paris.
179
180 REFERENCES
[10] Courbot, A., 1996. Prevention of severe slugging in the dunbar 16 inchesmultiphase pipeline. In: Proceedings of the Annual Offshore Technol-ogy Conference. Vol. 4. pp. 445–452, SPE no. 8196.
[11] Devold, H., 2010. Oil and gas production handbook, An introductionto oil and gas production, 2nd Edition. ABB Oil and Gas, Oslo.
[12] Di Meglio, F., 2011. Production de petrole: etude dynamique etcontrole des ecoulements a bouchons. Phd. thesis, l’Ecole des minesde paris.
[13] Di Meglio, F., Kaasa, G.-O., Petit, N., 2009. A first principle model formultiphase slugging flow in vertical risers. In: Joint 48th IEEE Con-ference on Decision and Control and 28th Chinese Control Conference.Shanghai, China, pp. 8244–8251.
[14] Di Meglio, F., Kaasa, G.-O., Petit, N., Alstad, V., 2010. Reproducingslugging oscillations of a real oil well. In: 49th IEEE Conference onDecision and Control. Atlanta, Georgia, USA, pp. 4473–4479.
[15] Di Meglio, F., Kaasa, G.-O., Petit, N., Alstad, V., 2012. Model-basedcontrol of slugging: advances and challenges. In: IFAC Workshop onAutomatic Control in Offshore Oil and Gas Production. Trondheim,Norway, pp. 109–115.
[16] Doyle, J., Glover, K., Khargonekar, P., Francis, B., aug 1989. State-space solutions to standard H2 and H∞ control problems. IEEE Trans-actions on Automatic Control 34 (8), 831–847.
[17] Eikrem, G. O., 2008. Eikrem riser model.URL http://www.nt.ntnu.no/users/skoge/diplom/prosjekt08/tuvnes/
[18] Eikrem, G. O., Aamo, O. M., Siahaan, H., Foss, B., 2004. Anti-slugcontrol of gas-lift wells - experimental results. In: 6th IFAC Symposiumon Nonlinear Control Systems. Stuttgart, Germany, pp. 1–8.
[19] Eikrem, G. O., Foss, B., Imsland, L., Hu, B., Golan, M., 2002. Stabi-lization of gas lifted wells. In: 15th IFAC World Congress. Barcelona,Spain, pp. 1489–1489.
[20] Eikrem, G. O., Imsland, L., Foss, B., 2004. Stabilization of gas liftedwells based on state estimation. In: IFAC International Symposium onAdvanced Control of Chemical Processes. Hong Kong, China, pp. 1–6.
181
[21] Fard, M. P., Godhavn, J.-M., Sagatun, S. I., 2006. Modelling of severeslug and slug control with olga, spe 84685. SPE Journal of Production& Operations.
[22] Glover, K., 1986. Robust stabilization of linear multivariable systems:relations to approximation. International Journal of Control 43 (3),741–766.
[23] Glover, K., Doyle, J. C., 1988. State-space formulae for all stabilizingcontrollers that satisfy an H∞-norm bound and relations to relationsto risk sensitivity. Systems and Control Letters 11 (3), 167–172.
[24] Glover, K., McFarlane, D., 1989. Robust stabilization of normalizedcoprime factor plant descriptions with h∞-bounded uncertainty. IEEETransactions on Automatic Control 34 (8), 821–830.
[25] Godhavn, J.-M., Fard, M. P., Fuchs, P. H., 2005. New slug controlstrategies, tuning rules and experimental results. Journal of ProcessControl 15, 547–557.
[26] Haaland, S. E., 1983. Simple and explicit formulas for the friction factorin turbulent pipe flow. Journal of Fluids Engineering 105 (1), 89–90.
[27] Havre, K., Skogestad, S., 1998. Effect of rhp zeros and poles on the sen-sitivity functions in multivariable systems. Journal of Process Control8 (3), 155–164.
[28] Havre, K., Skogestad, S., 2001. Achievable performance of multivari-able systems with unstable zeros and poles. International Journal ofControl 48, 1131–1139.
[29] Havre, K., Skogestad, S., August 2003. Selection of variables for sta-bilizing control using pole vectors. IEEE Transaction on AutomaticControl 48 (8), 1393–1398.
[30] Havre, K., Stornes, K., Stray, H., 2000. Taming slug flow in pipelines.ABB Review 4, 55–63.
[31] Hu, B., Golan, M., 2003. Gas-lift instability resulted production lossand its remedy by feedback control: dynamical simulation results. In:SPE International Improved Oil Recovery Conference in Asia Pacific.SPE no. 84917-MS, Kuala Lumpur, Malaysia, pp. 513–521.
182 REFERENCES
[32] Irmann-Jacobsen, T. B., 2012. Flow assurance - a system perspective.MEK4450 Offshore Technology Course, universitet i Oslo, MatematiskInstitutt.
[33] Isidori, A., 1999. Nonlinear Control Systems II, 1st Edition. Springer-Verlag, London.
[34] Jahanshahi, E., Skogestad, S., Auguest 2011. Simplified dynamicalmodels for control of severe slugging in multiphase risers. In: 18thIFAC World Congress. Milan, Italy, pp. 1634–1639.
[35] Jahanshahi, E., Skogestad, S., 2013. Closed-loop model identificationand pid/pi tuning for robust anti-slug control. In: 10th IFAC Inter-national Symposium on Dynamics and Control of Process Systems.Mumbai, India.
[36] Jahanshahi, E., Skogestad, S., 2013. Comparison between nonlin-ear modelbased controllers and gain-scheduling internal model controlbased on identified model. In: 52nd IEEE Conference on Decision andControl. Florence, Italy.
[37] Jahanshahi, E., Skogestad, S., Grøtli, E. I., 2013. Anti-slug control ex-periments using nonlinear observer. In: American Control Conference.Washington D.C., pp. 1056–1062.
[38] Jahanshahi, E., Skogestad, S., Grøtli, E. I., 2013. Nonlinear model-based control of two-phase flow in risers by feedback linearization. In:9th IFAC Symposium on Nonlinear Control Systems. Toulouse, France,pp. 301–306.
[39] Jahanshahi, E., Skogestad, S., Hansen, H., 2012. Control structuredesign for stabilizing unstable gas-lift oil wells. In: Advanced Controlof Chemical Processes. Vol. 8. Singapore, pp. 93–100.
[40] Jahanshahi, E., Skogestad, S., Helgesen, A. H., May 2012. Controllabil-ity analysis of severe slugging in well-pipeline-riser systems. In: IFACWorkshop - Automatic Control in Offshore Oil and Gas Production.Trondheim, Norway, pp. 101–108.
[41] Jahanshahi, E., Skogestad, S., Lieungh, M., 2013. Subsea solution foranti-slug control of multiphase risers. In: European Control Conference.Zurich, Switzerland, pp. 4094–4099.
183
[42] Jankovic, M., Sepulchre, R., Kokotovic, P., dec 1996. Constructive Lya-punov stabilization of nonlinear cascade systems. IEEE Transactions onAutomatic Control 41 (12), 1723 –1735.
[43] Jansen, B., Dalsmo, M., Nøkleberg, L., Havre, K., Kristiansen, V.,Lemetayer, P., 1999. Automatic control of unstable gas lifted wells. In:SPE Annual Technical Conference and Exhibition. Houston, Texas, pp.1–9.
[44] Jansen, F., Shoham, O., Taitel, Y., 1996. The elimination of severeslugging - experiments and modeling. International Journal of Multi-phase Flow 22 (6), 1055 – 1072.
[45] Johal, K., 2012. Flow Assurance for Oil-Gas Fields Production Trans-port, fisrt Edition. Fluids in Motion Limited, London.
[46] Joshi, N. B., Muhammad, M., Creek, J., McFadden, J., 2003. Flowassurance: A challenging path to well completions and productivity.In: Offshore Technology Conference. Houston, Texas, pp. 1142–1146.
[47] Julier, S. J., Uhlmann, J. K., 1997. A new extension of the kalman filterto non-linear systems. In: AeroSense: The 11th Int. Symp. A.D.S.S.C.
[48] Kaasa, G.-O., Alstad, V., Zhou, J., Aamo, O. M., 2007. Nonlinearmodel-based control of unstable wells. Journal of Modeling, Identifica-tion and Control 28 (3), 69–79.
[49] Kaasa, G.-O., Alstad, V., Zhou, J., Aamo, O. M., 2008. Attenuationof slugging in unstable oil wells by nonlinear control. In: 17th IFACWorld Congress. COEX, Korea, South, pp. 6251–6256.
[50] Kalman, R. E., Falb, P. L., Arbib, M. A., 1969. Topics in mathematicalsystem theory. McGraw-Hill Book Co., New York.
[51] Khalil, H. K., 2002. Nonlinear Systems, Third Edition. Prentice Hall,Upper Saddle River, New Jersey.
[52] Krima, H., Cao, Y., Lao, L., May 2012. Gas injection for hydrodynamicslug control. In: IFAC Workshop, Automatic Control in Offshore Oiland Gas Production. Vol. 1. Trondheim, Norway, pp. 116–121.
[53] Larsen, R., Lund, A., Argo, C., June 2003. Cold flow - a practicalsolution. In: 11th International Conference on Multiphase Flow. SanRemo,Italy, pp. 313–331.
184 REFERENCES
[54] Lieungh, M., June 2012. Stabilizing slug control using sub-sea chokevalve. Master thesis, Norwegian University of Science and Technology,Deoartment of Chemical Engineering.
[55] Martins Da Silva, C., Dessen, F., Nydal, O. J., 2010. Dynamic multi-phase flow models for control. In: BHR Group - 7th North AmericanConference on Multiphase Technology. pp. 221–236.
[56] Meglio, F. D., Petit, N., Alstad, V., Kaasa, G.-O., 2012. Stabilization ofslugging in oil production facilities with or without upstream pressuresensors. Journal of Process Control 22 (4), 809 – 822.
[57] Mokhatab, S., Poe, W. A., Speight, J. G., 2006. Chapter 3 - raw gastransmission. In: Handbook of Natural Gas Transmission and Process-ing. Gulf Professional Publishing, Burlington, pp. 81 – 188.
[58] Morari, M., Zafiriou, E., 1989. Robust Process Control. Prentice Hall,Englewood Cliffs, New Jersey.
[59] Nilsen, A. S., 2012. Simplified first principle model for severe-sluggingflow in s-shaped risers. Master project, Nowegian University of Scienceand Technology.
[60] Ogazi, A., Cao, Y., Yeung, H., Lao, L., 2010. Slug control with largevalve openings to maximize oil production. SPE Journal 15 (3), 812–821.
[61] Ogazi, A., Ogunkolade, S., Cao, Y., Lao, L., Yeung, H., 2009. Severeslugging control through open loop unstable pid tuning to increase oilproduction. pp. 17–32.
[62] Ogazi, I., Cao, Y., Lao, L., Yeung, H., 2011. Production potential ofsevere slugging control systems. In: 18th IFACWorld Congress. Vol. 18.Milano, Italy, pp. 10869–10874.
[63] Pickering, P. F., Hewitt, G. F., Watson, M. J., Hale, C. P., June 2001.The prediction of flows in production risers - truth & myth? In: IIRConference. Aberdeen, pp. 1–16.
[64] Saberi, A., Kokotovic, P., Sussmann, H., dec 1989. Global stabilizationof partially linear composite systems. In: Proceedings of the 28th IEEEConference on Decision and Control, 1989. pp. 1385 –1391 vol.2.
185
[65] Sayda, A., Taylor, J., 2007. Modeling and control of three-phase grav-ilty separators in oil production facilities. In: American Control Con-ference, 2007. ACC ’07. pp. 4847–4853.
[66] Schlumberger, 2013. The schlumberger oilfield glossary.http://www.glossary.oilfield.slb.com.
[67] Schmidt, Z., Brill, J., Beggs, D., 1980. Riser-base gas injection into these forties line. SPE Journal 20, 407–414.
[68] Schmidt, Z., Brill, J. P., Beggs, H. D., 1979. Choking can eliminatesevere pipeline slugging. Oil & Gas Journal 12, 230–238.
[69] Sinegre, L., 2006. Etude des instabilits dans les puits activs par gas-lift.Ph.D. thesis, Ecole des Mines de Paris.
[70] Sivertsen, H., 2008. Stabilizing of desired flow regimes using activecontrol. Ph.D. thesis, Norwegian University of Science and Technology,NTNU.
[71] Sivertsen, H., Alstad, V., Skogestad, S., 2009. Medium-scale experi-ments on stabilizing riser-slug flow. SPE Projects, Facilities & Con-struction 4 (4), 156–170, SPE no. 120040.
[72] Skofteland, G., Godhavn, J. M., Kulset, T., 2007. Implementation ofa slug control system for subsea wells in an integrated operation envi-ronment. In: 13th International Conference on Multiphase ProductionTechnology. Edinburgh, UK., pp. 225–236.
[73] Skogestad, S., 2009. Chemical And Energy Process Engineering. CRCPress, Taylor & Francis Group, Boca Raton, FL.
[74] Skogestad, S., Postlethwaite, I., 2005. Multivariable Feedback Control:Analysis and Design. Wiley & Sons, Chichester, West Sussex, UK.
[75] Storkaas, E., 2005. Stabilizing control and controllability: Control solu-tions to avoid slug flow in pipeline-riser system. Ph.D. thesis, NorwegianUniversity of Science and Technology, NTNU.
[76] Storkaas, E., Skogestad, S., 2003. A low-dimensional dynamic model ofsevere slugging for control design and analysis. In: 11th InternationalConference on Multiphase flow. BHR Group, San Remo, Italy, pp. 117–133.
186 REFERENCES
[77] Storkaas, E., Skogestad, S., 2007. Controllability analysis of two-phasepipeline-riser systems at riser slugging conditions. Control EngineeringPractice 15 (5), 567–581.
[78] Taitel, Y., 1986. Stability of severe slugging. International Journal ofMultiphase Flow 12 (2), 203–217.
[79] Time, R. W., May 2011. Flow assurance and multiphase flow. Seminarat Aker Solutions, Stavanger.
[80] Tuvnes, H., 2008. Final year project: Severe well slugging and reservoir-well interactions, modeling and simulations. Tech. rep., Norwegian Uni-versity of Science and Techology.
[81] Vorotnikov, V. I., 1997. Partial Stability and Control, 1st Edition.Birkhauser, Boston.
[82] Yocum, B., April 1973. Offshore riser slug flow avoidance: Mathemati-cal models for design and optimization. In: SPE European Meeting. So-ciety of Petroleum Engineers, London, United Kingdom, p. SPE 4312.
[83] Yuwana, M., Seborg, D. E., May 1982. A new method for on-line con-troller tuning. AIChE Journal 28 (3), 434–440.